0:00.000–0:02.852
You're in the right place if you use ChatGPT,
0:02.852–0:05.936
Claude, Claude Code, Claude Co-work, or Codex,
0:05.936–0:09.800
but don't know which one you should use for business and how.
0:09.800–0:13.840
Because in this video, we're talking about using AI agents effectively.
0:13.840–0:16.430
Specifically, we're talking about Codex,
0:16.430–0:17.725
OpenAI's agent tool,
0:17.725–0:19.308
which is like Claude Code,
0:19.308–0:24.129
except it's an easy-to-use interface for doing knowledge work and coding work.
0:24.129–0:29.031
In this video, we're going to be breaking down the seven core capabilities of Codex,
0:29.031–0:30.711
OpenAI's AI agent super app,
0:30.711–0:32.602
each with real-world use cases,
0:32.602–0:37.924
so that by the end, you know exactly [music] what it can do and where it fits in your workflow.
0:37.924–0:39.853
Let's get it fits in your workflow.
0:39.853–0:40.880
Let's get started.
0:40.880–0:42.551
So, this right here is Codex.
0:42.551–0:47.961
This is a clean interface for AI agents that can fully control your computer.
0:47.961–0:51.738
At first glance, it kind of looks like ChatGPT or Claude,
0:51.738–0:53.748
because you can create a chat,
0:53.748–0:55.998
but they're not exactly the same.
0:55.998–0:59.271
When you chat on ChatGPT or Claude and upload files,
0:59.271–1:02.164
those files are actually stored in the cloud.
1:02.164–1:10.267
With Codex, all of the files that you upload to Codex or that you create with Codex are stored on your computer.
1:10.267–1:14.880
Codex is made to do any task that you might want to do on a computer.
1:14.880–1:17.480
You can create a motion graphics video.
1:17.480–1:19.666
You can create beautiful landing pages,
1:19.666–1:24.103
and directly inside Codex, you can generate images that you can place in your site.
1:24.103–1:29.280
On Codex, you can also create more complex games and 3D simulations.
1:29.280–1:34.464
You can also create mobile apps with a front end and back end with a professional design.
1:34.464–1:38.056
You can also create desktop apps in just a few prompts.
1:38.056–1:42.554
You can ask GPT 5.5 to create you an Excel sheet after doing a ton of research,
1:42.554–1:45.767
and it can create spreadsheets with multiple slides.
1:45.767–1:50.779
It can also create any type of Word doc with many different charts,
1:50.779–1:59.254
and you can also create presentations directly inside Codex and export them to Canva or any other tool you want.
1:59.254–2:03.804
And I haven't even mentioned yet that Codex can fully control my computer,
2:03.804–2:06.005
and it can also control the browser.
2:06.005–2:08.248
But, this is something we'll talk about a little bit later.
2:08.248–2:12.315
As we go through the main seven different capabilities of Codex,
2:12.315–2:14.951
you'll realize that it is truly a super app
2:14.951–2:16.533
that does any coding task
2:16.533–2:18.265
or any knowledge work task.
2:18.265–2:19.723
Let's not waste any time.
2:19.723–2:22.012
Let's hop into capability number one.
2:22.012–2:26.089
So, the first capability that Codex has is full file access.
2:26.089–2:28.375
You see, when I use ChatGPT,
2:28.375–2:31.456
all of the files that I give to ChatGPT
2:31.456–2:35.640
or that ChatGPT generates are stored in the cloud.
2:35.640–2:37.337
However, when we use Codex,
2:37.337–2:39.328
all of the files that are created
2:39.328–2:42.795
or the files that I give to it are stored on your computer.
2:42.795–2:48.299
And the agent inside Codex has full access to your computer.
2:48.299–2:49.339
Let me show you.
2:49.339–2:51.150
In my downloads folder,
2:51.150–2:54.409
I have a folder called receipts to process.
2:54.409–2:57.386
And we just have 60 different little photos of receipts
2:57.386–2:59.069
that I've added to this folder.
2:59.069–3:00.317
I can simply ask Codex,
3:00.317–3:01.762
"In the downloads folder,
3:01.762–3:04.258
you'll see my receipts that I need to process.
3:04.258–3:07.250
I want you to analyze them in an Excel sheet
3:07.250–3:10.242
and create charts that helps me visualize
3:10.242–3:12.123
the types of transactions
3:12.123–3:15.030
and any other patterns that you notice."
3:15.030–3:16.996
So, we can choose our model,
3:16.996–3:18.535
which is GPT 5.5 high,
3:18.535–3:19.732
and we can run it.
3:19.732–3:20.616
And the agent,
3:20.616–3:22.606
because it has full file access,
3:22.606–3:24.891
it can access any file on my computer,
3:24.891–3:26.807
it'll find the relevant folder
3:26.807–3:28.207
and complete the task.
3:28.207–3:29.359
And you can see here,
3:29.359–3:30.918
it found the target folder,
3:30.918–3:34.171
which is users Riley Brown downloads receipts to process
3:34.171–3:35.459
with only 53 receipts.
3:35.459–3:36.888
I'm going to OCR those,
3:36.888–3:40.575
and so we're going to extract the text from those receipts,
3:40.575–3:43.358
classify and categorize each transaction.
3:43.358–3:47.470
Now, it's creating an Excel workbook on my computer
3:47.470–3:49.680
that we can open up in Codex.
3:49.680–3:50.266
All right.
3:50.266–3:51.700
So, after 7 minutes,
3:51.700–3:52.629
it is now done
3:52.629–3:56.341
and we can very easily just open this Excel project up
3:56.341–3:57.944
directly inside Code X
3:57.944–4:01.066
and you can see that it has created this file.
4:01.066–4:02.839
We can see there's a dashboard,
4:02.839–4:05.090
the total spend is there's a dashboard,
4:05.090–4:06.600
the total spend is 25,982.
4:06.600–4:07.699
We see receipts,
4:07.699–4:08.955
category summary,
4:08.955–4:10.133
payment summary,
4:10.133–4:12.724
and we can see all the different trends,
4:12.724–4:14.608
the spend by payment method.
4:14.608–4:16.344
We can see the monthly trends.
4:16.344–4:17.809
This is an Excel sheet
4:17.809–4:20.087
and because it's on our computer,
4:20.087–4:22.040
if you click this down arrow,
4:22.040–4:23.748
you can see open in folder
4:23.748–4:28.142
and we can see the exact file location of this Excel spreadsheet.
4:28.142–4:32.870
So, now that we know that all of the files created on Code X live on your computer,
4:32.870–4:35.736
the question becomes how do you stay organized?
4:35.736–4:41.407
And I will tell you the best way that you can stay organized inside Code X is to work in a project.
4:41.407–4:44.280
So, you can create a chat in one of two ways.
4:44.280–4:50.558
One way you can create a chat is if you just go to this chats category and you just hit this new chat button.
4:50.558–4:54.674
It allows you to create a chat that just shows up here in the sidebar.
4:54.674–4:57.449
It's not actually inside a project.
4:57.449–5:04.022
Or, you can go ahead and you can create a project and you can always hit start at existing folder.
5:04.022–5:11.012
all of my projects in my documents folder and then I will just create a new project seven capabilities
5:11.012–5:13.591
cuz that's what we're talking about.
5:13.591–5:20.349
We can create a new folder and now we see that we're creating a new chat inside the seven capabilities folder.
5:20.349–5:26.328
You'll see here this project will show up here in the left side panel and as soon as I add a chat,
5:26.328–5:28.502
you can see the chat shows up here.
5:28.502–5:34.909
The cool part about Code X is you can multitask and you can create multiple chats at the same time.
5:34.909–5:40.000
Here, I'm going to ask the agent to create a Word doc.
5:40.000–5:44.790
And we're doing this within a chat inside the seven capabilities folder.
5:44.800–5:47.245
Any document that the agent creates
5:47.245–5:48.467
when it's in a chat
5:48.467–5:49.608
inside a project
5:49.608–5:52.950
will actually store the document that it creates
5:52.950–5:55.150
in this folder on your computer.
5:55.160–5:57.667
So, if we click these three dots here
5:57.667–5:59.172
and say open in finder
5:59.172–6:02.013
and we open the seven categories folder,
6:02.013–6:04.604
we'll see that there's nothing in it.
6:04.604–6:07.638
And that's because the agent is not done creating the document.
6:07.638–6:09.058
As soon as the agent's done,
6:09.058–6:10.972
it's going to put as the agent's done,
6:10.972–6:14.306
it's going to put the uh document that it creates in this project.
6:14.306–6:16.139
And as you can see here,
6:16.139–6:19.130
we have this Word doc that was created
6:19.130–6:23.955
and it's called learn 90% of OpenAI Codex in under 2 minutes.
6:23.955–6:26.602
You can see the title of the folder right here.
6:26.602–6:30.068
If we go ahead and open up finder at that location
6:30.068–6:32.408
and we click on outputs and docs,
6:32.408–6:36.221
you can see that that document is located right here.
6:36.221–6:40.188
So, this is literally on my computer by default.
6:40.188–6:45.122
And so, this video that I'm making now is called the seven capabilities of Codex.
6:45.122–6:48.852
And if we were to create a new chat in this folder,
6:48.852–6:53.702
the chat would be aware that it is starting out in this project.
6:53.702–6:55.653
I can type at learn and see
6:55.653–7:02.620
I can actually at mention any file that's created within this seven capabilities folder.
7:02.620–7:04.483
And I can say,
7:04.483–7:09.902
"Please turn this into a landing page,
7:09.902–7:15.840
a good learning resource for my viewers."
7:15.840–7:16.186
And so,
7:16.186–7:18.203
I can take any document within this folder
7:18.203–7:19.470
and I can just at mention it
7:19.470–7:19.874
and say,
7:19.874–7:21.660
"Please turn this to a landing page."
7:21.660–7:23.504
And because it can control my computer
7:23.504–7:26.280
and go look at all of the documents that are organized very
7:26.280–7:26.711
easily,
7:26.711–7:28.251
it can find the right document
7:28.251–7:29.853
and then create a landing page.
7:29.853–7:33.332
And the code for that landing page will exist on my computer.
7:33.332–7:34.840
And here we go, it's done.
7:34.840–7:41.919
We can open this up and we can see that it created a landing page based on that document that is stored on our computer.
7:41.919–7:44.729
And let's go ahead and move on to the next capability.
7:44.729–7:48.527
Next up, Codex has something called persistent memory.
7:48.527–7:50.812
So, within Codex, there's two types of memory.
7:50.812–7:52.505
There's manual memory,
7:52.505–7:54.283
and there's auto memory.
7:54.283–7:57.752
Manual memory is when you ask the agent to remember something,
7:57.752–7:58.687
when you tell it,
7:58.687–8:01.889
"Hey, can you remember this going forward in the future?"
8:01.889–8:03.357
Let me show you an example.
8:03.357–8:03.564
So,
8:03.564–8:05.838
I made a few changes to that landing page
8:05.838–8:07.905
and got it to create it in this format
8:07.905–8:09.697
with this little left sidebar.
8:09.697–8:10.945
We can scroll down,
8:10.945–8:14.688
and we can automatically snap to the section that we like.
8:14.688–8:16.190
I really like this format,
8:16.190–8:16.873
and we like.
8:16.873–8:18.886
I really like this format,
8:18.886–8:20.259
and so from now on,
8:20.259–8:25.017
can you please use this format whenever I ask for a landing page
8:25.017–8:27.680
and I don't specify uh the styling?
8:27.680–8:29.724
You should use this styling
8:29.724–8:35.599
and remember that if there's a lot of text content to use this table of contents.
8:35.599–8:35.852
So,
8:35.852–8:39.230
I can simply just ask the agent to remember this,
8:39.230–8:43.536
and it's automatically going to add it to the agents.md file,
8:43.536–8:45.731
which is where you can see here,
8:45.731–8:47.335
search for agents.md.
8:47.335–8:50.865
This is where it's going to store this information for later.
8:50.865–8:51.333
And so,
8:51.333–8:52.424
you can see here,
8:52.424–8:54.840
the agent updated its manual memory,
8:54.840–8:56.710
which is the agents.md file.
8:56.710–8:57.767
If we click on this,
8:57.767–9:00.806
here are the things that I've asked it to add in the past.
9:00.806–9:02.773
Here's the what it just added.
9:02.773–9:06.658
It says for landing pages that are made for learning purposes,
9:06.658–9:08.825
it says use this text first format,
9:08.825–9:12.112
um and then it kind of describes the style that I like.
9:12.112–9:18.216
This agents.md file is a living document that gets updated over time
9:18.216–9:21.000
that acts as the manual memory.
9:21.000–9:25.373
But, there's also a different type of memory that you should never
9:25.373–9:30.859
and this is an auto memory feature that Codex kind of keeps updated automatically.
9:30.859–9:33.197
And just like the agents.md file,
9:33.197–9:36.287
this is all stored in a file on your computer.
9:36.287–9:41.513
If we go to Codex and hit command N to create a new chat,
9:41.513–9:45.432
please tell me about what you have in the
9:45.432–9:48.757
and then Codex / memories folder.
9:48.757–9:53.516
I want to understand what you know.
9:53.516–9:56.031
Also, open that file.
9:56.031–10:00.106
And you can see here it gave me the path to this file.
10:00.106–10:02.525
We can go ahead and open this up.
10:02.525–10:05.214
And this is a document you should not touch.
10:05.214–10:10.234
You can just kind of observe it to see what the agent observes about you.
10:10.234–10:13.720
And so every few weeks I'll actually go through this and take a look.
10:13.720–10:19.380
And this actually gives me a nice little summary of all of the different tasks that I've asked my agent to do.
10:19.380–10:23.711
And it can be useful when helping other people build out similar workflows.
10:23.711–10:27.760
I do not recommend making manual changes to these memories, right?
10:27.760–10:31.043
Just let the agent store the memories however it wants to
10:31.043–10:34.186
because the agent has a skill that keeps these updated
10:34.186–10:36.072
and it will get better over time.
10:36.072–10:39.457
The agents.md file on the other hand is your manual memory.
10:39.457–10:41.520
This is meant for you to keep it updated.
10:41.520–10:43.540
You can tell the agent to update it
10:43.540–10:45.993
or you can manually update it if you want.
10:45.993–10:50.720
The third huge capability that Codex has is plugins.
10:50.720–10:54.004
And so plugins are basically just Codex connections.
10:54.004–11:02.856
They are installable, reusable bundles that connect the OpenAI Codex AI agent to external tools, apps, and workflows.
11:02.856–11:05.940
And this allows them to have other functionality
11:05.940–11:09.174
and they can connect to all of the tools that you use.
11:09.174–11:11.806
And so if you go to the plugins tab,
11:11.806–11:14.155
with two clicks I can add Slack
11:14.155–11:15.659
and I can add Gmail.
11:15.659–11:17.495
Once you've added Gmail,
11:17.495–11:22.216
and you can see here this plugin actually comes with some skills.
11:22.216–11:23.866
So there's a Gmail skill
11:23.866–11:27.497
and an inbox triage skill that is part of this plugin.
11:27.497–11:28.053
Right now,
11:28.053–11:30.276
I think there's over 100 different plugins
11:30.276–11:31.882
that allow you to connect Codex
11:31.882–11:33.611
to the tools that you already use.
11:33.611–11:33.814
So,
11:33.814–11:35.233
let's go over an example.
11:35.233–11:35.427
So,
11:35.427–11:36.722
I created a new project,
11:36.722–11:39.507
which is a folder on my computer called brand deals,
11:39.507–11:40.607
and I'm going to say,
11:40.607–11:43.327
"I want you to look at my email from the past 2 weeks.
11:43.327–11:46.271
Find all the brands offering to do paid promotion,
11:46.271–11:47.813
do research on all of them,
11:47.813–11:50.406
and put them in a table with relevant notes."
11:50.406–11:52.158
And then I can type the at sign,
11:52.158–11:53.770
and here I can type in Gmail.
11:53.770–11:56.278
All of the plugins show up in this menu,
11:56.278–11:59.177
so we can very easily at mention the plugins.
11:59.177–12:02.799
And this is a quick way to see which plugins that you have access to.
12:02.799–12:05.483
And I really like using this menu,
12:05.483–12:08.560
and then we can go ahead and click send.
12:08.560–12:12.144
And now Codex will search through all of my recent emails
12:12.144–12:14.965
because I've given it access via the plugin,
12:14.965–12:16.566
and it will get back to us.
12:16.566–12:19.604
This is taking about like 5 minutes so far,
12:19.604–12:21.948
and I have another idea for a task
12:21.948–12:23.597
I want an AI agent to do.
12:23.597–12:23.798
So,
12:23.798–12:26.619
I can very easily just hit this new message button,
12:26.619–12:28.701
and I can type out exactly what I want.
12:28.701–12:29.023
So,
12:29.023–12:33.440
I'm making my next video on CLIs and MCPs in Codex.
12:33.440–12:35.083
Please look at my Notion,
12:35.083–12:36.804
and so I at mention Notion,
12:36.804–12:37.977
which is a plugin,
12:37.977–12:40.960
and find my high-quality long-form scripts.
12:40.960–12:43.845
Then I want you to please write a script on this topic.
12:43.845–12:45.920
Use Excalidraw diagrams.
12:45.920–12:46.907
This is a skill,
12:46.907–12:48.360
which I'll get to next.
12:48.360–12:49.657
We can run this prompt.
12:49.657–12:52.204
And so now we have two AI agents working,
12:52.204–12:53.980
both working with plugins.
12:53.980–12:54.229
So,
12:54.229–12:56.225
the brand deals task is done.
12:56.225–12:59.676
If we go ahead and open up this document right here,
12:59.676–13:01.812
we can see that it created file,
13:01.812–13:04.195
and it went through all of my emails
13:04.195–13:09.371
and wrote down the companies offering to sponsor my videos all the way down.
13:09.371–13:11.911
It literally went through all of them
13:11.911–13:13.550
and took detailed notes
13:13.550–13:14.205
and said,
13:14.205–13:15.844
"What did they ask for?"
13:15.844–13:18.630
And also they added some research notes.
13:18.630–13:21.467
And then I could ask it to respond to the emails
13:21.467–13:24.527
because of the same way that it can read my emails,
13:24.527–13:26.991
it can also send emails from my account.
13:26.991–13:29.751
And if we go to the Notion plugin that we tested,
13:29.751–13:32.440
we can see here that it created this document,
13:32.440–13:34.209
and it went through my Notion,
13:34.209–13:36.898
found all the scripts that I made previously,
13:36.898–13:37.606
and it said,
13:37.606–13:41.710
"Do you ever feel like AI coding tools are getting more powerful every
13:41.720–13:44.160
week, but also somehow more confusing?"
13:44.160–13:46.055
And this is literally in my voice
13:46.055–13:48.510
because it went through all of my scripts,
13:48.510–13:51.668
and it created this document using the Notion plugin.
13:51.668–13:55.082
And that brings us to the fourth capability that Codex has,
13:55.082–13:56.127
which are skills.
13:56.127–14:02.003
And you can think of skills on Codex as reusable workflow recipes or SOPs
14:02.003–14:06.410
that your agent can use many times over and over again.
14:06.410–14:09.018
In Codex, if you go up to the plugins tab,
14:09.018–14:12.653
you see that there's actually a subtab within plugins,
14:12.653–14:13.838
which are skills.
14:13.838–14:18.025
And in the skills tab, these are similar to plugins.
14:18.025–14:20.042
They're more like instruction files
14:20.042–14:24.012
that your agent will follow every time it wants to use one of these skills.
14:24.012–14:27.095
So, all your skills can be found in the plugins folder
14:27.095–14:28.566
in the skills subfolder.
14:28.566–14:31.190
But, how do you actually create a skill?
14:31.190–14:34.049
In order to create a skill, you can type a prompt just like this.
14:34.049–14:39.171
I want you to create a skill called brand deal analysis that does blank.
14:39.171–14:42.611
And I could enter this prompt in and create this skill,
14:42.611–14:44.293
and then once that's done,
14:44.293–14:47.580
in order to use any skill, you just press this slash.
14:47.580–14:53.390
And so, I can use my Excalidraw diagram skill just by use my Excalidraw diagram skill just by typing slash.
14:53.390–14:55.573
I can also use my YouTube researcher skill
14:55.573–14:57.195
or my YouTube thumbnail skill,
14:57.195–14:58.941
which we'll get to in just a skill,
14:58.941–15:00.760
which we'll get to in just a second.
15:00.760–15:04.382
These are all skills that I can quickly access by typing slash.
15:04.382–15:06.567
These are different than plugins,
15:06.567–15:08.526
which I can access by typing at.
15:08.526–15:12.600
So, there's two main ways to create a skill in Codex.
15:12.600–15:14.361
One is the prompt to skill,
15:14.361–15:19.726
which is basically just I want you to create a skill called blank that does blank.
15:19.726–15:22.878
This will give you a reusable instructions file,
15:22.878–15:23.955
which is a skill,
15:23.955–15:27.040
but I don't think it will be as good as option one.
15:27.040–15:29.633
The best way to create a skill, in my opinion,
15:29.633–15:31.806
is to use the manual workflow method,
15:31.806–15:34.820
where you basically ask your agent to do something,
15:34.820–15:35.521
it'll do it,
15:35.521–15:38.325
and then you basically go back and forth, right?
15:38.325–15:39.798
Tell it to create something,
15:39.798–15:41.761
and say, "Hey, I want you to change it."
15:41.761–15:43.602
And you say, "Get the output right."
15:43.602–15:45.565
If you're not happy, you iterate more,
15:45.565–15:47.836
and you kind of go back through this process,
15:47.836–15:50.658
and you loop around until you're happy with the output.
15:50.658–15:52.843
And then you reverse engineer the skill,
15:52.843–15:55.671
and you will basically end up saying something like,
15:55.671–15:57.214
"I'm happy with this output.
15:57.214–15:58.502
Turn it into a skill."
15:58.502–16:00.147
And I'll actually show you.
16:00.147–16:01.784
If we go back to Codex,
16:01.784–16:03.329
and you can see here,
16:03.329–16:09.330
I've asked the agent to put all those paid promos into a spreadsheet right here.
16:09.330–16:11.508
And I really like this spreadsheet.
16:11.508–16:12.880
I think it's high quality.
16:12.880–16:15.023
I think it's color-coded, right,
16:15.023–16:16.784
by the different priority.
16:16.784–16:18.367
This is looking really good.
16:18.367–16:19.098
And remember,
16:19.098–16:20.559
you can always full screen it
16:20.559–16:24.040
to get a better look at whatever document it is that you're creating.
16:24.040–16:24.428
And so,
16:24.428–16:25.720
well, all we need to do is
16:25.720–16:27.206
we need to just type this in.
16:27.206–16:28.600
"I'm happy with this output.
16:28.600–16:30.936
Turn this into a skill that I can use."
16:30.936–16:33.422
And the agent is smart enough to be like,
16:33.422–16:38.772
"Okay, this is he wants to always be able to analyze his Gmail for brand sponsorships.
16:38.772–16:39.961
Alternate into a skill."
16:39.961–16:40.301
And so,
16:40.301–16:40.980
now it's done.
16:40.980–16:41.572
It says,
16:41.572–16:47.920
"I've created and validated the reusable skill called brand deal researcher skill.md."
16:47.920–16:50.286
And we can open up this skill by uh
16:50.286–16:52.564
clicking on the skill.md file,
16:52.564–16:55.105
and we can actually read this over.
16:55.105–16:59.231
And it turned our entire workflow into these instructions.
16:59.231–17:03.618
And the AI is very smart enough at taking anything that you've done with your agent,
17:03.618–17:05.081
and turning it into a skill.
17:05.081–17:05.370
So,
17:05.370–17:08.738
now if we hit command N to create a new chat,
17:08.738–17:11.432
"I want you uh to please do {slash}
17:11.432–17:14.223
It's called uh to please do {slash}
17:14.223–17:17.120
It's called brand deal researcher.
17:17.120–17:17.509
Right?
17:17.509–17:17.754
So,
17:17.754–17:21.748
now we have this reusable skill that we can use at any point.
17:21.748–17:22.761
And you can see here
17:22.761–17:24.596
it's using all of the same plugins.
17:24.596–17:26.710
It's using the Gmail plugin.
17:26.710–17:26.955
So,
17:26.955–17:30.878
a skill can contain instructions to use a specific plugin
17:30.878–17:32.757
because at the end of the day
17:32.757–17:35.699
it's just instructions to do a simple task.
17:35.699–17:39.085
While our research brand deal task is loading,
17:39.085–17:45.423
I actually went ahead and did a create Excalidraw graphics for the seven capabilities.
17:45.423–17:46.810
The video that we're doing now,
17:46.810–17:48.943
remember this document that we created earlier
17:48.943–17:53.797
which is this landing page that describes all of the different things that we're talking about in this video.
17:53.797–17:55.582
I went ahead and I just said,
17:55.582–18:00.937
"Please look at that document and make a ton of graphics on a single Excalidraw doc.
18:00.937–18:05.755
Use the Excalidraw diagram skill and make the chart super useful."
18:05.755–18:11.960
So, then the AI thought and worked for 4 minutes and 23 seconds and then it created this.
18:11.960–18:13.816
And we can open this up and we can just
18:13.816–18:19.261
And we can open this up and we can just automatically basically import this project and boom, look at this.
18:19.261–18:23.360
It created this entire outline and I love this formatting.
18:23.360–18:27.320
This formatting that it did with the background here, I actually really like that.
18:27.320–18:30.357
And so, it created all of the diagrams.
18:30.357–18:32.204
Wow, this is actually fantastic.
18:32.204–18:33.672
It has the intro,
18:33.672–18:35.559
all of the summaries,
18:35.559–18:36.293
memory,
18:36.293–18:38.390
plugins versus skills,
18:38.390–18:39.858
skill creation.
18:39.858–18:40.944
Very cool.
18:40.944–18:42.765
So, now I can go back to Codex
18:42.765–18:47.121
and I can say, "Wow, I really like the formatting that you did here.
18:47.121–18:48.550
Please update the skills
18:48.550–18:51.748
so that you always put them in the containers that you did
18:51.748–18:53.790
and I want you to always give me a link
18:53.790–18:56.036
with all of the diagrams that you create
18:56.036–18:58.417
when I ask you to create multiple diagrams
18:58.417–18:59.302
just like this.
18:59.302–19:00.910
I love it."
19:00.920–19:02.769
And so, even once you've used a skill,
19:02.769–19:04.319
maybe you've used it for a while
19:04.319–19:05.035
and you like it
19:05.035–19:08.911
and then you get surprised by something valuable like this diagram right here
19:08.911–19:10.223
or these this formatting,
19:10.223–19:12.320
you can always just say, "Hey, I want you to
19:12.320–19:13.998
always format it like this."
19:13.998–19:18.821
And so, every time you use the skill is an opportunity to make the skill even better.
19:18.821–19:21.944
Just tell your agent to update the skill to always do something.
19:21.944–19:27.356
The next capability of Codex is built-in GPT image to access.
19:27.356–19:33.038
So, you can generate images with the best image model in the world directly inside Codex.
19:33.038–19:35.994
So, I created a new project called content,
19:35.994–19:41.003
and here I'm saying, "Generate a product photo for my uh sweater company.
19:41.003–19:43.296
This is our first sweater.
19:43.296–19:45.199
I can paste an image of a sweater,
19:45.199–19:46.398
and you can see here,
19:46.398–19:47.948
this is the first sweater,
19:47.948–19:49.922
knitted sweater that I'm making.
19:49.922–19:53.506
Please create five images with GPT/image/2
19:53.506–19:55.153
of different models
19:55.153–19:57.865
of different of different models
19:57.865–20:00.190
of different nationalities
20:00.190–20:02.030
wearing this sweater.
20:02.030–20:05.587
Three of the images should have one person,
20:05.587–20:08.551
one image should have three people,
20:08.551–20:11.920
the last image should have five people."
20:11.920–20:13.411
And so, we can run this,
20:13.411–20:15.451
and within the content folder,
20:15.451–20:17.334
it's going to create images.
20:17.334–20:20.064
And those images will live inside the content folder
20:20.064–20:20.870
once it's done.
20:20.870–20:21.653
And look at this.
20:21.653–20:26.136
You can see here that it's using the skill.md for image gen.
20:26.136–20:29.500
And so, this image gen skill is built into Codex.
20:29.500–20:32.052
And so, you can go to Codex, click on skills,
20:32.052–20:35.242
and you can see which skills are built into the system.
20:35.242–20:37.375
And you don't have to enable them,
20:37.375–20:40.118
but these are recommended skills by Codex,
20:40.118–20:41.880
and one of them is image gen.
20:41.880–20:45.218
And by default, this uses the latest and best image model.
20:45.218–20:48.481
And you can see here that it generated one of the images.
20:48.481–20:50.320
We still need to wait for the other four.
20:50.320–20:52.827
Okay, here, it just generated the second image.
20:52.827–20:54.727
It just generated the third image.
20:54.727–20:57.823
The next image should have three people in one image.
20:57.823–20:58.840
And here we go.
20:58.840–21:00.179
Here is number four,
21:00.179–21:01.754
and here is number five.
21:01.754–21:04.871
And now that you know that you can generate images,
21:04.871–21:07.764
that brings me to the next capability of Codex,
21:07.764–21:10.139
which is browser use and computer use.
21:10.139–21:13.396
And so in order to use computer use, it is a plugin.
21:13.396–21:16.273
So remember, you just type in at computer use.
21:16.273–21:20.079
Now I can say, "Please open the Canva app on my computer,
21:20.079–21:21.816
make a new presentation,
21:21.816–21:25.539
and put each one of these images on each of the slides."
21:25.539–21:27.111
So one image per slide,
21:27.111–21:29.510
there should be five total slides.
21:29.520–21:32.085
We can simply tell Codex to control our computer
21:32.085–21:32.983
and it will do it.
21:32.983–21:34.686
And as you can see here,
21:34.686–21:36.120
we see list Mac apps
21:36.120–21:38.361
and now it's looking at Canva.
21:38.361–21:39.502
You can see here,
21:39.502–21:41.622
it's controlling my computer.
21:41.622–21:42.931
I'm not even touching it.
21:42.931–21:45.006
It just created this presentation.
21:45.006–21:46.746
It's clicking in Canva.
21:46.746–21:47.234
Okay.
21:47.234–21:49.444
Oh, it just ran a script,
21:49.444–21:50.438
I believe.
21:50.438–21:52.400
And so it just ran a script.
21:52.400–21:54.303
It's looking at Canva
21:54.303–21:55.994
and you can see here
21:55.994–21:58.636
it created this presentation
21:58.636–22:02.336
and it has all five images on the computer.
22:02.336–22:03.565
And you can see here,
22:03.565–22:04.649
this is the mouse.
22:04.649–22:08.302
You can see the mouse outline of the computer use.
22:08.302–22:11.081
It will move around from time to time.
22:11.081–22:14.360
It created this presentation for me.
22:14.360–22:16.305
So this right here is a browser
22:16.305–22:17.549
and you can see that
22:17.549–22:19.572
if we literally exit out of this
22:19.572–22:20.738
and hit plus sign,
22:20.738–22:21.983
we can see browser.
22:21.983–22:24.461
And when we open this index.html file,
22:24.461–22:27.171
it's literally opening this in a browser.
22:27.171–22:29.027
So now I'm going to say,
22:29.027–22:30.882
"Turn this into an app,
22:30.882–22:32.836
then test the interface
22:32.836–22:37.914
and make sure the buttons and navigation work using at browser
22:37.914–22:41.440
and navigation work using at browser use."
22:41.440–22:43.161
And I can run this prompt here.
22:43.161–22:45.060
And now it's going to take this,
22:45.060–22:46.229
turn it into an app,
22:46.229–22:48.274
and then it's just going to test it
22:48.274–22:50.840
by controlling the browser inside Codex.
22:50.840–22:52.004
Okay, you can see here,
22:52.004–22:53.842
it's kind of like a different color.
22:53.842–22:55.985
Look, this is my mouse right here.
22:55.985–22:57.596
This is the browser use.
22:57.596–23:00.295
So the AI is actually controlling the browser
23:00.295–23:02.496
and it's going to start testing them.
23:02.496–23:05.285
So, we see that it's test start button,
23:05.285–23:08.596
and the AI is now taking control of the screen.
23:08.596–23:11.450
The first visible check showed the app controls,
23:11.450–23:15.626
but the start button smooth scroll is too subtle for live verification.
23:15.626–23:17.425
You can see here it's moving around.
23:17.425–23:19.720
It's scrolling through the app.
23:19.720–23:27.120
It's clicking on the different quizzes that it's created.
23:27.120–23:29.088
Now, it's testing quiz completion,
23:29.088–23:31.515
and now it just marked this a correct answer.
23:31.515–23:41.080
So, it's making sure that all parts of the application work.
23:41.080–23:42.975
Oh, now it tested the side panel,
23:42.975–23:44.600
and now we're at the bottom.
23:44.600–23:45.592
This is very cool.
23:45.592–23:48.725
Anything you can open up in the browser,
23:48.725–23:50.623
you can ask browser use.
23:50.623–23:54.595
You can use the @browseruse plugin
23:54.595–23:58.440
to test it directly inside Codex.
23:58.440–24:02.846
And finally, capability number seven inside Codex is automation.
24:02.846–24:06.407
You can create automation simply by talking to Codex.
24:06.407–24:09.868
So, remember this Excel sheet that we created in this chat here
24:09.868–24:13.595
in the brand deals project called research paid promo brand deals.
24:13.595–24:18.452
We created this spreadsheet after it analyzed all of my incoming emails
24:18.452–24:20.000
with the email plugin.
24:20.000–24:21.479
Then, we took this output,
24:21.479–24:23.093
and we turned it into a skill.
24:23.093–24:24.902
I'm happy with this output.
24:24.902–24:26.920
Please turn it into a skill.
24:26.920–24:29.947
Now, what I can do is I can repeatedly do this
24:29.947–24:32.469
on a weekly or daily or hourly basis.
24:32.469–24:40.575
I can simply say, "Please do this every single Friday at 9:00 a.m., please,
24:40.575–24:43.200
and update this sheet."
24:43.200–24:46.269
What this is going to do is it's going to create an automation.
24:46.269–24:49.400
Notice here we only see two automations.
24:49.400–24:52.428
Here it says, "I'll set this up as a recurring automation for Fridays at 9:00
24:52.428–24:54.360
recurring automation for Fridays at 9:00 a.m."
24:54.360–24:56.040
And we can see it directly in the UI.
24:56.040–24:58.806
Created weekly brand deal sheet updated
24:58.806–25:02.711
and if we click on this, we can see the automation details.
25:02.711–25:05.242
We can see that at Friday at 9:00 a.m.
25:05.242–25:07.022
We can see that it's active.
25:07.022–25:09.273
We can see that the last run is empty
25:09.273–25:11.214
because it's never run before
25:11.214–25:15.250
and we can see that we are using the brand deal researcher skill
25:15.250–25:20.762
to scan Gmail for paid promotion sponsorship brand deal influencer campaigns, etc.
25:20.762–25:21.711
So created this
25:21.711–25:24.046
and now if we go to the automations tab,
25:24.046–25:25.286
we now see it up here.
25:25.286–25:26.835
We can see Friday at 9:00 a.m.
25:26.835–25:28.964
We have another one at Fridays at 9:00 a.m.
25:28.964–25:30.087
We can click on this
25:30.087–25:31.772
and you can edit it directly.
25:31.772–25:35.037
And finally, we have a bonus capability inside Codex
25:35.037–25:37.412
which is something called Chronicle.
25:37.412–25:39.723
So this is a brand new feature within Codex.
25:39.723–25:40.972
If we go to settings
25:40.972–25:42.377
and we click settings
25:42.377–25:44.251
and we go to personalization
25:44.251–25:45.500
and we scroll down,
25:45.500–25:49.404
you will see an option to to use Chronicle research preview.
25:49.404–25:50.660
When you turn this on,
25:50.660–25:53.381
what this will do is it'll say that it is running
25:53.381–25:55.892
and this one may seem a little bit invasive.
25:55.892–25:59.578
What this one does is it basically is always recording your screen
25:59.578–26:02.125
so it has context with what you're working on.
26:02.125–26:04.142
And so if I go to my presentation here
26:04.142–26:06.697
and I just kind of scroll around in my browser,
26:06.697–26:08.916
use kind of scroll around in my browser,
26:08.916–26:09.800
use Chronicle.
26:09.800–26:18.440
Uh tell me what I should add to my Codex visual presentation.
26:18.440–26:22.951
And so it shows you here all use Chronicle skill for live screen work content
26:22.951–26:24.680
then cross everything over.
26:24.680–26:26.074
Chronicle's available
26:26.074–26:28.652
so I'm checking the fresh screen state first
26:28.652–26:29.628
and look at this,
26:29.628–26:33.680
you can see that it's pulling up recent screenshots of my screen here.
26:33.680–26:36.146
And so it is pulling old images
26:36.146–26:41.967
and it's getting context from what I've had open on my screen recently.
26:41.967–26:45.781
And so it's able to keep track of all the things that I'm working on
26:45.781–26:49.040
and the way that you invoke it is just say use Chronicle.
26:49.040–26:50.691
And so it says based on Chronicle,
26:50.691–26:52.814
I'd add these to your visual presentation.
26:52.814–26:54.383
A Codex super app map,
26:54.383–26:57.697
a full control loop diagram for prompt files,
26:57.697–26:59.440
a Chronicle demo slide,
26:59.440–27:01.184
reusable skills slide,
27:01.184–27:03.887
and why Codex over other tools slide.
27:03.887–27:05.271
And so, it automatically,
27:05.271–27:08.101
and I didn't give it any [clears throat] information,
27:08.101–27:09.359
I didn't upload an image
27:09.359–27:11.183
or give it an Excalidraw document,
27:11.183–27:13.520
it just knew what was in this slide deck here.
27:13.520–27:15.458
And it's able to understand it
27:15.458–27:18.715
because this Chronicle feature watches my screen
27:18.715–27:21.041
and takes screenshots constantly.
27:21.041–27:24.953
And so, those are the main features and abilities within Codex.
27:24.953–27:28.126
And in summary, we covered that Codex has full file control,
27:28.126–27:30.792
meaning you can create, edit, and delete any file.
27:30.792–27:32.117
It has persistent memory,
27:32.117–27:36.030
and whether that is automatic memory or manual memory in your agents.md file.
27:36.030–27:37.640
We also talked about plugins,
27:37.640–27:39.507
which are built into the platform.
27:39.507–27:42.587
It connects your agent to all of the tools that exist.
27:42.587–27:47.074
You can create skills by taking any task that you create with the agent
27:47.074–27:49.560
and then converting that into a skill.
27:49.560–27:53.015
Your Codex agent has access to image generation,
27:53.015–27:56.808
and it is the best image generation model in the world,
27:56.808–27:59.589
which is generation model in the world,
27:59.589–28:01.200
which is GPT-image-2.
28:01.200–28:03.796
We also have browser control and computer use,
28:03.796–28:07.400
and this allows your agent to control your computer like a human.
28:07.400–28:08.582
The human interface,
28:08.582–28:10.880
which is a keyboard, a mouse, and a screen,
28:10.880–28:12.259
can now be used by agents,
28:12.259–28:14.491
and it's getting better exponentially.
28:14.491–28:15.771
And we also have automations.
28:15.771–28:18.006
You can take any useful task that your agent does
28:18.006–28:19.515
and say, "Hey, do this every day.
28:19.515–28:20.712
Hey, do this every hour.
28:20.712–28:21.825
Hey, do this every week.
28:21.825–28:23.618
Do this every month." That type of thing.
28:23.618–28:25.860
And then as a bonus, we talked about Chronicle.
28:25.860–28:28.206
Anyway, thank you guys so much for watching.
28:28.206–28:30.035
I hope you got a lot out of this video.
28:30.035–28:32.074
I make videos like this every single week,
28:32.074–28:34.520
so make sure to subscribe and hit that like button.
28:34.520–28:35.310
It helps me out a lot.
28:35.310–28:38.840
I'll see you here for the next video.
0:00.000–0:02.852
如果你使用 ChatGPT，
0:02.852–0:05.936
Claude、Claude Code、Claude Co-work 或 Codex，
0:05.936–0:09.800
但不知道該在商業上如何使用它們，
0:09.800–0:13.840
因為在這個影片中，我們將探討如何有效使用 AI 代理。
0:13.840–0:16.430
因為在本影片中，我們將探討如何有效使用 AI 代理。
0:16.430–0:17.725
具體來說，我們將討論 Codex，
0:17.725–0:19.308
OpenAI 的代理工具，
0:19.308–0:24.129
它是一個易於使用的介面，用於處理知識工作和程式碼工作。
0:24.129–0:29.031
但提供了一個易於使用的介面，用於進行知識工作和程式碼編寫。
0:29.031–0:30.711
在本影片中，我們將拆解 Codex 的七項核心功能，
0:30.711–0:32.602
OpenAI 的 AI 代理超級應用程式，
0:32.602–0:37.924
這樣到最後，你會清楚知道 [音樂] 它能做什麼，以及它如何融入你的工作流程。
0:37.924–0:39.853
讓你在看完後，能清楚知道 [音樂] 它能做什麼，以及它如何融入你的工作流程。
0:39.853–0:40.880
讓它融入你的工作流程。
0:40.880–0:42.551
所以，這裡就是 Codex。
0:42.551–0:47.961
這是一個乾淨的介面，讓 AI 代理能夠完全控制你的電腦。
0:47.961–0:51.738
這是一個乾淨的 AI 代理介面，可以完全控制你的電腦。
0:51.738–0:53.748
乍看之下，它有點像 ChatGPT 或 Claude，
0:53.748–0:55.998
因為你可以建立聊天，
0:55.998–0:59.271
當你在 ChatGPT 或 Claude 上聊天並上傳檔案時，
0:59.271–1:02.164
這些檔案實際上儲存在雲端。
1:02.164–1:10.267
使用 Codex 時，你上傳到 Codex 或使用 Codex 建立的檔案都會儲存在你的電腦上。
1:10.267–1:14.880
Codex 旨在執行你在電腦上可能想進行的任何任務。
1:14.880–1:17.480
你可以建立動態圖形影片。
1:17.480–1:19.666
你可以建立精美的落地頁，
1:19.666–1:24.103
並且直接在 Codex 內生成圖片，然後將其放置在你的網站上。
1:24.103–1:29.280
在 Codex 上，你也可以建立更複雜的遊戲和 3D 模擬。
1:29.280–1:34.464
你也可以建立具有前端和後端且具備專業設計的行動應用程式。
1:34.464–1:38.056
你也可以僅透過幾個提示詞建立桌面應用程式。
1:38.056–1:42.554
你可以要求 GPT 5.5 在完成大量研究後為你建立 Excel 試算表，
1:42.554–1:45.767
並且它能建立包含多個工作表的試算表。
1:45.767–1:50.779
它也能建立包含多種不同圖表的任何類型 Word 文件，
1:50.779–1:59.254
你也可以直接在 Codex 內建立簡報，並將其匯出到 Canva 或你想要的任何其他工具。
1:59.254–2:03.804
我甚至還沒提到 Codex 可以完全控制我的電腦，
2:03.804–2:06.005
而且它也能控制瀏覽器。
2:06.005–2:08.248
但這是我們稍後會討論的內容。
2:08.248–2:12.315
當我們逐步了解 Codex 的七項主要功能時，
2:12.315–2:14.951
你會發現它確實是一個超級應用程式，
2:14.951–2:16.533
能夠執行任何程式碼任務
2:16.533–2:18.265
或任何知識性工作任務。
2:18.265–2:19.723
我們別浪費時間了。
2:19.723–2:22.012
讓我們直接進入第一個功能。
2:22.012–2:26.089
所以，Codex 的第一個功能是擁有完整的檔案存取權限。
2:26.089–2:28.375
你看，當我使用 ChatGPT 時，
2:28.375–2:31.456
我提供給 ChatGPT 的所有檔案
2:31.456–2:35.640
或是 ChatGPT 產生的檔案，都會儲存在雲端。
2:35.640–2:37.337
然而，當我們使用 Codex 時，
2:37.337–2:39.328
所有產生的檔案
2:39.328–2:42.795
或是我提供給它的檔案，都會儲存在你的電腦上。
2:42.795–2:48.299
而且 Codex 內部的代理程式可以完全存取你的電腦。
2:48.299–2:49.339
讓我展示給你看看。
2:49.339–2:51.150
在我的下載資料夾中，
2:51.150–2:54.409
我有一個名為「待處理收據」的資料夾。
2:54.409–2:57.386
裡面只有 60 張不同的收據小照片
2:57.386–2:59.069
是我新增到這個資料夾裡的。
2:59.069–3:00.317
我只需要簡單地詢問 Codex，
3:00.317–3:01.762
「在下載資料夾裡，
3:01.762–3:04.258
你會看到我需要處理的收據。
3:04.258–3:07.250
我希望你將它們分析成 Excel 試算表
3:07.250–3:10.242
並建立圖表，幫助我視覺化
3:10.242–3:12.123
交易類型
3:12.123–3:15.030
以及你注意到的任何其他模式。
3:15.030–3:16.996
所以，我們可以選擇我們的模型，
3:16.996–3:18.535
也就是 GPT 5.5 high，
3:18.535–3:19.732
然後執行它。
3:19.732–3:20.616
而這個代理程式，
3:20.616–3:22.606
因為它擁有完整的檔案存取權限，
3:22.606–3:24.891
它可以存取我電腦上的任何檔案，
3:24.891–3:26.807
它會找到相關的資料夾
3:26.807–3:28.207
並完成任務。
3:28.207–3:29.359
你可以在這裡看到，
3:29.359–3:30.918
它找到了目標資料夾，
3:30.918–3:34.171
也就是 users Riley Brown downloads receipts to process
3:34.171–3:35.459
裡面只有 53 張收據。
3:35.459–3:36.888
我將對這些收據進行 OCR，
3:36.888–3:40.575
因此我們將從這些收據中提取文字，
3:40.575–3:43.358
並對每筆交易進行分類和歸類。
3:43.358–3:47.470
現在，它正在我的電腦上建立一個 Excel 工作簿
3:47.470–3:49.680
我們可以在 Codex 中開啟它。
3:49.680–3:50.266
好的。
3:50.266–3:51.700
所以，經過 7 分鐘後，
3:51.700–3:52.629
現在已經完成了
3:52.629–3:56.341
我們可以非常輕鬆地直接在
3:56.341–3:57.944
Code X 內部開啟這個 Excel 專案
3:57.944–4:01.066
你可以看到它已經建立了這個檔案。
4:01.066–4:02.839
我們可以看到有一個儀表板，
4:02.839–4:05.090
儀表板上顯示總支出，
4:05.090–4:06.600
總支出為 25,982。
4:06.600–4:07.699
我們可以看到收據，
4:07.699–4:08.955
類別摘要，
4:08.955–4:10.133
付款摘要，
4:10.133–4:12.724
以及我們可以看見所有不同的趨勢，
4:12.724–4:14.608
按付款方式分列的支出。
4:14.608–4:16.344
我們可以看到每月的趨勢。
4:16.344–4:17.809
這是一個 Excel 試算表
4:17.809–4:20.087
而且因為它位於我們的電腦上，
4:20.087–4:22.040
如果你點擊這個向下箭頭，
4:22.040–4:23.748
你可以看到「在資料夾中開啟」
4:23.748–4:28.142
我們可以看見這個 Excel 試算表的確切檔案位置。
4:28.142–4:32.870
所以，現在我們知道在 Code X 上建立的所有檔案都保存在你的電腦上，
4:32.870–4:35.736
問題就變成瞭如何保持組織有序？
4:35.736–4:41.407
我會告訴你，在 Code X 中保持組織有序的最佳方式就是在專案中工作。
4:41.407–4:44.280
因此，你可以透過兩種方式建立聊天。
4:44.280–4:50.558
建立聊天的一種方式是，如果你直接進入這個聊天類別並點擊這個新聊天按鈕。
4:50.558–4:54.674
它允許你建立一個聊天，該聊天只會顯示在側邊欄中。
4:54.674–4:57.449
它實際上並不在專案內部。
4:57.449–5:04.022
或者，你可以繼續建立一個專案，並隨時點擊「開始於現有資料夾」。
5:04.022–5:11.012
我的文件資料夾中的所有專案，然後我將建立一個新的專案七種能力
5:11.012–5:13.591
因為我們正在討論的就是這個。
5:13.591–5:20.349
我們可以建立一個新資料夾，現在我們看到我們正在七種能力資料夾內建立新聊天。
5:20.349–5:26.328
你會看到這個專案會顯示在左側面板中，而一旦我新增聊天，
5:26.328–5:28.502
你就可以看到聊天顯示在這裡。
5:28.502–5:34.909
Code X 的酷之處在於你可以多工處理，並同時建立多個聊天。
5:34.909–5:40.000
在這裡，我將要求代理建立一個 Word 文件。
5:40.000–5:44.790
而我們是在七種能力資料夾內的聊天中進行此操作。
5:44.800–5:47.245
代理建立的任何文件
5:47.245–5:48.467
當它處於聊天中
5:48.467–5:49.608
在專案內部
5:49.608–5:52.950
實際上會儲存它所建立的文件
5:52.950–5:55.150
在你電腦上的這個資料夾中。
5:55.160–5:57.667
所以，如果我們點擊這裡的三個點
5:57.667–5:59.172
並選擇在訪客中開啟
5:59.172–6:02.013
然後開啟七個類別資料夾，
6:02.013–6:04.604
我們會發現裡面什麼都沒有。
6:04.604–6:07.638
這是因為代理程式尚未完成建立文件。
6:07.638–6:09.058
一旦代理程式完成工作，
6:09.058–6:10.972
它會顯示代理程式已完成，
6:10.972–6:14.306
它會將它在此專案中建立的文件放置於此。
6:14.306–6:16.139
正如你在這裡看到的，
6:16.139–6:19.130
我們有一個已建立的 Word 文件
6:19.130–6:23.955
名為「兩分鐘內學會 OpenAI Codex 90% 的內容」。
6:23.955–6:26.602
你可以直接在這裡看到資料夾的標題。
6:26.602–6:30.068
如果我們繼續在該位置開啟訪客
6:30.068–6:32.408
並點擊輸出和文件，
6:32.408–6:36.221
你可以看到該文件位於這裡。
6:36.221–6:40.188
所以，這實際上預設就在我的電腦上。
6:40.188–6:45.122
因此，我現在製作的這部影片名為「Codex 的七項能力」。
6:45.122–6:48.852
如果我們要在這個資料夾中建立新的對話，
6:48.852–6:53.702
對話會知道它是在這個專案中開始的。
6:53.702–6:55.653
我可以輸入 @learn 並查看
6:55.653–7:02.620
我實際上可以@提及在這個七項能力資料夾中建立的任何檔案。
7:02.620–7:04.483
我可以說，
7:04.483–7:09.902
「請將這個轉成登入頁面，
7:09.902–7:15.840
作為我觀眾的良好學習資源。」
7:15.840–7:16.186
因此，
7:16.186–7:18.203
我可以取用這個資料夾中的任何文件，
7:18.203–7:19.470
然後直接@提及它，
7:19.470–7:19.874
並說，
7:19.874–7:21.660
「請將這個轉成登入頁面。」
7:21.660–7:23.504
因為它能控制我的電腦，
7:23.504–7:26.280
並去查看所有組織得非常好的文件，
7:26.280–7:26.711
輕鬆地，
7:26.711–7:28.251
它能找到正確的檔案，
7:28.251–7:29.853
然後建立登入頁面。
7:29.853–7:33.332
該登入頁面的程式碼將存在於我的電腦上。
7:33.332–7:34.840
好了，完成了。
7:34.840–7:41.919
我們可以打開它，看到它根據儲存在我們電腦上的那份文件建立了一個登入頁面。
7:41.919–7:44.729
讓我們繼續進行下一個能力。
7:44.729–7:48.527
接下來，Codex 擁有一項稱為持久記憶的功能。
7:48.527–7:50.812
所以，在 Codex 中有兩種類型的記憶。
7:50.812–7:52.505
有手動記憶，
7:52.505–7:54.283
以及自動記憶。
7:54.283–7:57.752
手動記憶是指你要求代理程式記住某些內容，
7:57.752–7:58.687
當你告訴它，
7:58.687–8:01.889
「嘿，你能記住這個，以便在未來使用嗎？」
8:01.889–8:03.357
讓我舉個例子。
8:03.357–8:03.564
所以，
8:03.564–8:05.838
我對那個登陸頁面做了一些修改，
8:05.838–8:07.905
並讓它以這種格式建立，
8:07.905–8:09.697
帶有這個小小的左側邊欄。
8:09.697–8:10.945
我們可以向下滾動，
8:10.945–8:14.688
並且可以自動跳轉到我們喜歡的區塊。
8:14.688–8:16.190
我非常喜歡這種格式，
8:16.190–8:16.873
我們喜歡。
8:16.873–8:18.886
我非常喜歡這種格式，
8:18.886–8:20.259
所以從現在開始，
8:20.259–8:25.017
能否請你在我要求建立登陸頁面時，
8:25.017–8:27.680
且我沒有特別指定樣式時，
8:27.680–8:29.724
使用這種樣式，
8:29.724–8:35.599
並記住，如果有大量文字內容，請使用這個目錄。
8:35.599–8:35.852
所以，
8:35.852–8:39.230
我可以直接要求代理程式記住這個，
8:39.230–8:43.536
它會自動將它新增到 agents.md 檔案中，
8:43.536–8:45.731
你可以在這裡看到，
8:45.731–8:47.335
搜尋 agents.md。
8:47.335–8:50.865
這就是它將此資訊儲存以供日後使用的地方。
8:50.865–8:51.333
因此，
8:51.333–8:52.424
你可以在這裡看到，
8:52.424–8:54.840
代理程式更新了它的手動記憶體，
8:54.840–8:56.710
也就是 agents.md 檔案。
8:56.710–8:57.767
如果我們點擊這裡，
8:57.767–9:00.806
這裡是我過去要求它新增的內容。
9:00.806–9:02.773
這是它剛剛新增的內容。
9:02.773–9:06.658
它指出，對於用於學習目的的登陸頁面，
9:06.658–9:08.825
它說使用這種文字優先格式，
9:08.825–9:12.112
嗯，然後它大致描述了我最喜歡的風格。
9:12.112–9:18.216
這個 agents.md 檔案是一個會隨時間更新的活文件，
9:18.216–9:21.000
它充當手動記憶體的角色。
9:21.000–9:25.373
但是，還有另一種類型的記憶體是你應該永遠不要
9:25.373–9:30.859
這是 Codex 會自動保持更新的自動記憶體功能。
9:30.859–9:33.197
就像 agents.md 檔案一樣，
9:33.197–9:36.287
這些資料都儲存在你電腦上的某個檔案裡。
9:36.287–9:41.513
如果我們打開 Codex 並按下 Command N 來建立新的對話，
9:41.513–9:45.432
請告訴我你在那個
9:45.432–9:48.757
在 Codex / memories 資料夾裡的內容。
9:48.757–9:53.516
我想了解你所知道的資訊。
9:53.516–9:56.031
同時，打開那個檔案。
9:56.031–10:00.106
你可以看到這裡顯示了該檔案的路徑。
10:00.106–10:02.525
我們可以繼續打開它。
10:02.525–10:05.214
這份文件你不應該去動它。
10:05.214–10:10.234
你只需要觀察它，看看 AI 代理者對你觀察到了什麼。
10:10.234–10:13.720
因此，我每隔幾週就會去檢查這份文件並查看內容。
10:13.720–10:19.380
這份文件實際上為我要求代理者執行的所有不同任務提供了一個簡潔的摘要。
10:19.380–10:23.711
在協助其他人建立類似的工作流程時，這會很有用。
10:23.711–10:27.760
我不建議手動修改這些記憶，對吧？
10:27.760–10:31.043
就讓代理者自行決定如何儲存記憶吧。
10:31.043–10:34.186
因為代理者具備讓這些記憶保持更新的技能。
10:34.186–10:36.072
而且它會隨著時間推移變得更好。
10:36.072–10:39.457
另一方面，agents.md 檔案是你的手動記憶。
10:39.457–10:41.520
這是要讓你來保持更新的。
10:41.520–10:43.540
你可以指示代理程式更新它
10:43.540–10:45.993
或者如果你願意，也可以手動更新它。
10:45.993–10:50.720
Codex 的第三項強大功能是外掛程式。
10:50.720–10:54.004
而外掛程式基本上就是 Codex 的連線。
10:54.004–11:02.856
它們是可安裝、可重複使用的套件，用於將 OpenAI Codex AI 代理程式與外部工具、應用程式和工作流程相連接。
11:02.856–11:05.940
這使得它們能夠具備其他功能
11:05.940–11:09.174
並且可以連接到你使用的所有工具。
11:09.174–11:11.806
因此，如果你前往外掛程式標籤頁，
11:11.806–11:14.155
只需兩下點擊，我就能新增 Slack
11:14.155–11:15.659
以及新增 Gmail。
11:15.659–11:17.495
一旦你新增了 Gmail，
11:17.495–11:22.216
你可以在這裡看到這個外掛程式實際上附帶了一些技能。
11:22.216–11:23.866
所以這裡有 Gmail 技能
11:23.866–11:27.497
以及作為此外掛程式一部分的收件匣篩選技能。
11:27.497–11:28.053
目前，
11:28.053–11:30.276
我認為有超過 100 種不同的外掛程式
11:30.276–11:31.882
可以讓你將 Codex
11:31.882–11:33.611
連接到你已經使用的工具。
11:33.611–11:33.814
所以，
11:33.814–11:35.233
讓我們來看一個範例。
11:35.233–11:35.427
所以，
11:35.427–11:36.722
我建立了一個新專案，
11:36.722–11:39.507
也就是電腦上名為「品牌合作」的資料夾，
11:39.507–11:40.607
然後我會說，
11:40.607–11:43.327
「我想請你查看我過去兩週的電子郵件。
11:43.327–11:46.271
找出所有提供付費推廣的品牌，
11:46.271–11:47.813
對它們進行研究，
11:47.813–11:50.406
並將它們整理成包含相關備註的表格。」
11:50.406–11:52.158
接著我可以輸入 @ 符號，
11:52.158–11:53.770
然後在這裡輸入 Gmail。
11:53.770–11:56.278
所有外掛都會顯示在這個選單中，
11:56.278–11:59.177
因此我們可以非常輕鬆地以 @ 提及這些外掛。
11:59.177–12:02.799
這也是一種快速查看你可使用之外掛的方式。
12:02.799–12:05.483
我真的很喜歡使用這個選單，
12:05.483–12:08.560
接著我們可以繼續點擊發送。
12:08.560–12:12.144
現在 Codex 會搜尋我最近的所有電子郵件，
12:12.144–12:14.965
因為我已透過外掛授予它存取權限，
12:14.965–12:16.566
它會回覆我們。
12:16.566–12:19.604
到目前為止大約花了 5 分鐘，
12:19.604–12:21.948
我還有另一個想讓 AI 代理執行的任務想法
12:21.948–12:23.597
我希望由 AI 代理來執行。
12:23.597–12:23.798
所以，
12:23.798–12:26.619
我可以非常輕鬆地按下這個新訊息按鈕，
12:26.619–12:28.701
然後精確地輸入我想要的內容。
12:28.701–12:29.023
所以，
12:29.023–12:33.440
我正在製作關於 Codex 中 CLI 和 MCP 的下一部影片。
12:33.440–12:35.083
請查看我的 Notion，
12:35.083–12:36.804
然後我在這裡提到 Notion，
12:36.804–12:37.977
這是一個外掛程式，
12:37.977–12:40.960
並找到我高品質的長篇腳本。
12:40.960–12:43.845
然後我希望你針對這個主題撰寫腳本。
12:43.845–12:45.920
使用 Excalidraw 圖表。
12:45.920–12:46.907
這是一項技能，
12:46.907–12:48.360
我接下來會介紹。
12:48.360–12:49.657
我們可以執行這個提示。
12:49.657–12:52.204
現在我們有兩個 AI 代理正在運作，
12:52.204–12:53.980
兩者都使用外掛程式。
12:53.980–12:54.229
所以，
12:54.229–12:56.225
品牌合作任務已完成。
12:56.225–12:59.676
如果我們繼續打開這裡的文件，
12:59.676–13:01.812
我們可以看到它建立了檔案，
13:01.812–13:04.195
並且它瀏覽了我所有的電子郵件
13:04.195–13:09.371
並詳細記錄了所有提供贊助我影片的公司資訊。
13:09.371–13:11.911
它確實瀏覽了所有郵件
13:11.911–13:13.550
並做了詳細的筆記
13:13.550–13:14.205
並詢問，
13:14.205–13:15.844
「他們要求什麼？」
13:15.844–13:18.630
此外，它也加入了一些研究筆記。
13:18.630–13:21.467
接著，我可以要求它回覆電子郵件
13:21.467–13:24.527
因為它能讀取我的電子郵件，
13:24.527–13:26.991
它也能從我的帳號發送電子郵件。
13:26.991–13:29.751
如果我們前往之前測試過的 Notion 外掛，
13:29.751–13:32.440
我們可以看到這裡它建立了這份文件，
13:32.440–13:34.209
並且它瀏覽了我的 Notion，
13:34.209–13:36.898
找到了我之前製作的所有腳本，
13:36.898–13:37.606
並說道，
13:37.606–13:41.710
「你是否覺得 AI 編碼工具每星期都變得更強大，
13:41.720–13:44.160
但同時卻也變得更令人困惑？」
13:44.160–13:46.055
這確實是用我的語氣
13:46.055–13:48.510
因為它瀏覽了我所有的腳本，
13:48.510–13:51.668
並使用 Notion 外掛程式建立了這份文件。
13:51.668–13:55.082
這就帶我們來到 Codex 的第四項功能，
13:55.082–13:56.127
也就是技能（Skills）。
13:56.127–14:02.003
你可以將 Codex 中的技能視為可重複使用的工作流程配方或標準作業程序（SOP），
14:02.003–14:06.410
讓你的代理程式可以反覆多次使用。
14:06.410–14:09.018
在 Codex 中，如果你點擊上方的外掛程式（Plugins）標籤，
14:09.018–14:12.653
你會發現外掛程式標籤下其實有一個子標籤，
14:12.653–14:13.838
那就是技能（Skills）。
14:13.838–14:18.025
在技能標籤中，這些功能與外掛程式類似。
14:18.025–14:20.042
它們更像是指示檔案，
14:20.042–14:24.012
當你的代理程式想要使用其中一項技能時，它會遵循這些指示。
14:24.012–14:27.095
因此，你可以在外掛程式資料夾
14:27.095–14:28.566
的技能子資料夾中找到所有技能。
14:28.566–14:31.190
但是，你實際上該如何建立技能呢？
14:31.190–14:34.049
為了建立技能，你可以輸入這樣的提示詞。
14:34.049–14:39.171
我希望你建立一個名為「品牌合作分析」的技能，功能為空白。
14:39.171–14:42.611
我可以輸入這個提示詞並建立這個技能，
14:42.611–14:44.293
然後一旦完成，
14:44.293–14:47.580
要使用任何技能，你只需按下這個斜線鍵。
14:47.580–14:53.390
因此，我只需輸入斜線鍵，就能使用我的 Excalidraw 圖表技能，只要輸入斜線鍵即可使用我的 Excalidraw 圖表技能。
14:53.390–14:55.573
我也可以使用我的 YouTube 研究員技能
14:55.573–14:57.195
或是我的 YouTube 縮圖技能，
14:57.195–14:58.941
我們馬上就會談到這個技能，
14:58.941–15:00.760
我們馬上就會談到。
15:00.760–15:04.382
這些都是我可以透過輸入斜線快速存取的技能。
15:04.382–15:06.567
這些與外掛程式不同，
15:06.567–15:08.526
外掛程式我可以透過輸入 at 符號來存取。
15:08.526–15:12.600
因此，在 Codex 中建立技能主要有兩種方式。
15:12.600–15:14.361
一種是提示轉技能，
15:14.361–15:19.726
基本上就是「我想要你建立一個名為空白、執行空白功能的技能」。
15:19.726–15:22.878
這會給你一個可重複使用的指示檔案，
15:22.878–15:23.955
也就是技能，
15:23.955–15:27.040
但我認為它不會比選項一好。
15:27.040–15:29.633
我認為建立技能的最佳方式，
15:29.633–15:31.806
是使用手動工作流程方法，
15:31.806–15:34.820
基本上就是要求你的代理程式執行某項任務，
15:34.820–15:35.521
它會執行，
15:35.521–15:38.325
然後你基本上會來回調整，對吧？
15:38.325–15:39.798
告訴它建立某樣東西，
15:39.798–15:41.761
然後說：「嘿，我希望你修改它。」
15:41.761–15:43.602
然後你說：「把輸出結果弄對。」
15:43.602–15:45.565
如果你不滿意，就繼續迭代，
15:45.565–15:47.836
然後你大致上會重新走過這個流程，
15:47.836–15:50.658
並循環這個過程，直到你對輸出結果滿意為止。
15:50.658–15:52.843
接著你逆向工程這項技能，
15:52.843–15:55.671
你基本上最終會說出類似下面的話：
15:55.671–15:57.214
「我對這個輸出結果很滿意。
15:57.214–15:58.502
把它變成一項技能。」
15:58.502–16:00.147
我實際上會展示給你。
16:00.147–16:01.784
如果我們回到 Codex，
16:01.784–16:03.329
你可以在這裡看到，
16:03.329–16:09.330
我已經要求代理程式將所有付費促銷內容放入這裡的試算表中。
16:09.330–16:11.508
我非常喜歡這個試算表。
16:11.508–16:12.880
我認為它的品質很高。
16:12.880–16:15.023
我認為它已經依不同優先級進行了色彩編碼，對吧？
16:15.023–16:16.784
依不同的優先級。
16:16.784–16:18.367
這看起來真的很不錯。
16:18.367–16:19.098
並且記住，
16:19.098–16:20.559
你隨時可以將其全螢幕顯示
16:20.559–16:24.040
以便更清楚地查看你正在建立的任何文件。
16:24.040–16:24.428
所以，
16:24.428–16:25.720
我們只需要
16:25.720–16:27.206
輸入這些內容。
16:27.206–16:28.600
「我對這個輸出結果很滿意。
16:28.600–16:30.936
把它變成一項我可以使用的技能。」
16:30.936–16:33.422
而這個 AI 代理足夠聰明，會理解說，
16:33.422–16:38.772
「好的，他總是希望能夠分析他的 Gmail 來尋找品牌贊助。
16:38.772–16:39.961
將其轉換為一項技能。」
16:39.961–16:40.301
所以，
16:40.301–16:40.980
現在已經完成了。
16:40.980–16:41.572
它顯示，
16:41.572–16:47.920
「我已建立並驗證了一個名為 brand deal researcher skill.md 的可重用技能。」
16:47.920–16:50.286
我們可以透過點擊
16:50.286–16:52.564
skill.md 檔案來開啟這個技能，
16:52.564–16:55.105
並且我們實際上可以閱讀這些內容。
16:55.105–16:59.231
它將我們整個工作流程轉化為這些指示。
16:59.231–17:03.618
而 AI 非常聰明，能夠將你與代理互動的任何內容，
17:03.618–17:05.081
轉化為一項技能。
17:05.081–17:05.370
所以，
17:05.370–17:08.738
現在如果我們按下 Command N 來建立新的對話，
17:08.738–17:11.432
「我希望你呃，請執行 {slash}
17:11.432–17:14.223
這叫做呃，請執行 {slash}
17:14.223–17:17.120
這叫做品牌合作研究員。
17:17.120–17:17.509
對吧？
17:17.509–17:17.754
所以，
17:17.754–17:21.748
現在我們擁有一個可重複使用的技能，可以在任何時候使用。
17:21.748–17:22.761
你可以在這裡看到
17:22.761–17:24.596
它使用了所有相同的插件。
17:24.596–17:26.710
它使用了 Gmail 插件。
17:26.710–17:26.955
所以，
17:26.955–17:30.878
技能可以包含使用特定插件的指令
17:30.878–17:32.757
因為归根结底
17:32.757–17:35.699
它只是執行簡單任務的指令。
17:35.699–17:39.085
當我們的品牌合作研究任務載入時，
17:39.085–17:45.423
我實際上已經先行建立了七項能力的 Excalidraw 圖形。
17:45.423–17:46.810
我們現在正在製作的影片，
17:46.810–17:48.943
記得我們之前建立的這份文件嗎
17:48.943–17:53.797
也就是這份介紹了本影片中所討論各種不同內容的登陸頁面。
17:53.797–17:55.582
我已經先行說了，
17:55.582–18:00.937
「請查看那份文件，並在單一 Excalidraw 文件中製作大量圖形。
18:00.937–18:05.755
使用 Excalidraw 圖表技能，製作一張極具實用價值的圖表。
18:05.755–18:11.960
接著，AI 思考並運作了 4 分 23 秒，然後產生了這個結果。
18:11.960–18:13.816
我們可以打開它，然後直接
18:13.816–18:19.261
我們可以打開它，然後基本上自動匯入這個專案，哇，看看這個。
18:19.261–18:23.360
它建立了整個大綱，我喜歡這種格式。
18:23.360–18:27.320
它在這裡使用背景所做的格式，我其實非常喜歡。
18:27.320–18:30.357
因此，它建立了所有的圖表。
18:30.357–18:32.204
哇，這真的非常棒。
18:32.204–18:33.672
它有介紹，
18:33.672–18:35.559
所有的摘要，
18:35.559–18:36.293
記憶體，
18:36.293–18:38.390
外掛程式與技能的比較，
18:38.390–18:39.858
技能建立。
18:39.858–18:40.944
非常酷。
18:40.944–18:42.765
所以，現在我可以回到 Codex
18:42.765–18:47.121
然後我說：「哇，我真的很喜歡你這裡的格式。
18:47.121–18:48.550
請更新技能
18:48.550–18:51.748
以便你總是將它們放在你所做的容器中
18:51.748–18:53.790
我希望你總是給我一個連結
18:53.790–18:56.036
包含你所建立的所有圖表
18:56.036–18:58.417
當我要求你建立多個圖表時
18:58.417–18:59.302
就像這樣。
18:59.302–19:00.910
我非常喜歡。
19:00.920–19:02.769
因此，即使你已經使用過某項技能，
19:02.769–19:04.319
也許你已經使用它一段時間了
19:04.319–19:05.035
而且你喜歡它
19:05.035–19:08.911
然後你對像這裡這樣的圖表所帶來的價值感到驚訝
19:08.911–19:10.223
或是這種格式設定，
19:10.223–19:12.320
你總是可以說：「嘿，我希望你
19:12.320–19:13.998
總是這樣格式化它。」
19:13.998–19:18.821
因此，每次使用這項技能都是一次讓技能變得更好的機會。
19:18.821–19:21.944
只需告訴你的代理程式更新技能，以始終執行某項操作。
19:21.944–19:27.356
Codex 的下一項功能是內建的 GPT 圖像生成能力。
19:27.356–19:33.038
因此，你可以在 Codex 內部直接使用全球最佳的圖像模型來生成圖片。
19:33.038–19:35.994
因此，我建立了一個名為 content 的新專案，
19:35.994–19:41.003
這裡我說：「為我的毛衣公司生成一張產品照片。
19:41.003–19:43.296
這是我們的第一件毛衣。
19:43.296–19:45.199
我可以貼上一張毛衣的圖片，
19:45.199–19:46.398
你可以在這裡看到，
19:46.398–19:47.948
這是第一件毛衣，
19:47.948–19:49.922
我正在製作的針織毛衣。
19:49.922–19:53.506
請使用 GPT/image/2 建立五張圖片
19:53.506–19:55.153
由不同模特兒
19:55.153–19:57.865
不同模型的技能
19:57.865–20:00.190
來自不同國籍
20:00.190–20:02.030
穿著這件毛衣。
20:02.030–20:05.587
其中三張圖片應各有一人，
20:05.587–20:08.551
一張圖片應有三個人，
20:08.551–20:11.920
最後一張圖片應有五個人。”
20:11.920–20:13.411
所以，我們可以執行這個，
20:13.411–20:15.451
並在 content 資料夾中，
20:15.451–20:17.334
它會建立圖片。
20:17.334–20:20.064
一旦完成，這些圖片將存放在 content 資料夾內
20:20.064–20:20.870
完成後。
20:20.870–20:21.653
看看這個。
20:21.653–20:26.136
你可以在這裡看到它正在使用 skill.md 進行圖片生成。
20:26.136–20:29.500
因此，這個圖片生成技能已內建於 Codex 中。
20:29.500–20:32.052
所以，你可以前往 Codex，點擊技能，
20:32.052–20:35.242
並查看系統內建了哪些技能。
20:35.242–20:37.375
你不需要啟用它們，
20:37.375–20:40.118
但這些是 Codex 推薦的技能，
20:40.118–20:41.880
其中一項是圖像生成。
20:41.880–20:45.218
預設情況下，這會使用最新且最佳的圖像模型。
20:45.218–20:48.481
你可以在這裡看到它生成了一張圖片。
20:48.481–20:50.320
我們仍需等待其餘四張圖片。
20:50.320–20:52.827
好的，這裡它剛剛生成了第二張圖片。
20:52.827–20:54.727
它剛剛生成了第三張圖片。
20:54.727–20:57.823
下一張圖片應該會在一張圖片中包含三個人。
20:57.823–20:58.840
好了，我們繼續。
20:58.840–21:00.179
這是第四張，
21:00.179–21:01.754
這是第五張。
21:01.754–21:04.871
現在你已經知道可以生成圖片了，
21:04.871–21:07.764
這就引出了 Codex 的下一個功能，
21:07.764–21:10.139
也就是瀏覽器使用和電腦使用。
21:10.139–21:13.396
因此，要使用電腦使用功能，它是一個外掛程式。
21:13.396–21:16.273
所以請記住，你只需輸入 @computer use。
21:16.273–21:20.079
現在我可以說：「請在我的電腦上開啟 Canva 應用程式，
21:20.079–21:21.816
建立一份新的簡報，
21:21.816–21:25.539
並將這些圖片中的每一張放到每一張投影片上。」
21:25.539–21:27.111
所以每張投影片放一張圖片
21:27.111–21:29.510
總共應該有五張投影片。
21:29.520–21:32.085
我們可以簡單地告訴 Codex 來控制我們的電腦
21:32.085–21:32.983
它就會照做。
21:32.983–21:34.686
正如你在這裡看到的，
21:34.686–21:36.120
我們看到列出的 Mac 應用程式
21:36.120–21:38.361
現在它正在查看 Canva。
21:38.361–21:39.502
你可以在這裡看到，
21:39.502–21:41.622
它正在控制我的電腦。
21:41.622–21:42.931
我甚至都沒有碰它。
21:42.931–21:45.006
它剛剛建立了這份簡報。
21:45.006–21:46.746
它正在 Canva 裡點擊。
21:46.746–21:47.234
好的。
21:47.234–21:49.444
噢，它剛剛執行了一個腳本，
21:49.444–21:50.438
我相信是這樣。
21:50.438–21:52.400
所以它只是執行了一個腳本。
21:52.400–21:54.303
它正在查看 Canva
21:54.303–21:55.994
你可以在這裡看到
21:55.994–21:58.636
它建立了這份簡報
21:58.636–22:02.336
並且電腦裡有所有五張圖片。
22:02.336–22:03.565
你可以在這裡看到，
22:03.565–22:04.649
這是滑鼠。
22:04.649–22:08.302
你可以看到電腦使用的滑鼠游標。
22:08.302–22:11.081
它會不時地移動
22:11.081–22:14.360
它為我建立了這份簡報。
22:14.360–22:16.305
所以這裡是一個瀏覽器。
22:16.305–22:17.549
你可以看到
22:17.549–22:19.572
如果我們直接退出這個
22:19.572–22:20.738
並點擊加號，
22:20.738–22:21.983
我們可以看到瀏覽器。
22:21.983–22:24.461
當我們開啟這個 index.html 檔案時，
22:24.461–22:27.171
它實際上是在瀏覽器中開啟這個檔案。
22:27.171–22:29.027
所以現在我要說，
22:29.027–22:30.882
「將這個轉換成應用程式，
22:30.882–22:32.836
然後測試介面
22:32.836–22:37.914
並確保按鈕和導航功能在瀏覽器中使用正常
22:37.914–22:41.440
且導航功能在瀏覽器使用中運作正常。」
22:41.440–22:43.161
我可以執行這裡的提示。
22:43.161–22:45.060
現在它將處理這個，
22:45.060–22:46.229
將其轉換成應用程式，
22:46.229–22:48.274
然後它只會測試它
22:48.274–22:50.840
透過在 Codex 內部控制瀏覽器。
22:50.840–22:52.004
好的，你可以在這裡看到，
22:52.004–22:53.842
它的顏色有點不一樣。
22:53.842–22:55.985
看，這就是我的滑鼠指標。
22:55.985–22:57.596
這就是瀏覽器使用功能。
22:57.596–23:00.295
所以 AI 實際上正在控制瀏覽器
23:00.295–23:02.496
並且它將開始進行測試。
23:02.496–23:05.285
所以，我們看到它點擊了測試開始按鈕，
23:05.285–23:08.596
而 AI 現在正在掌控螢幕。
23:08.596–23:11.450
第一個可見的檢查顯示了應用程式的控制項，
23:11.450–23:15.626
但開始按鈕的平滑滾動對於即時驗證來說太微妙了。
23:15.626–23:17.425
你可以在這裡看到它正在移動。
23:17.425–23:19.720
它正在瀏覽應用程式。
23:19.720–23:27.120
它正在點擊它所建立的不同測驗。
23:27.120–23:29.088
現在，它正在測試測驗完成情況，
23:29.088–23:31.515
然後它剛才標記了這個正確答案。
23:31.515–23:41.080
所以，它正在確保應用程式的所有部分都能運作。
23:41.080–23:42.975
哦，現在它測試了側邊面板，
23:42.975–23:44.600
而我們現在已經到達底部。
23:44.600–23:45.592
這非常酷。
23:45.592–23:48.725
任何你可以在瀏覽器中開啟的內容，
23:48.725–23:50.623
都可以使用瀏覽器使用者功能。
23:50.623–23:54.595
你可以使用 @browseruse 外掛程式
23:54.595–23:58.440
直接在 Codex 內部進行測試。
23:58.440–24:02.846
最後，Codex 內的第七項功能是自動化。
24:02.846–24:06.407
你可以透過與 Codex 對話來簡單地建立自動化。
24:06.407–24:09.868
所以，請記住我們在這裡的對話中建立的這個 Excel 試算表，
24:09.868–24:13.595
在名為 research paid promo brand deals 的品牌合作專案中。
24:13.595–24:18.452
我們在使用電子郵件外掛程式分析了我所有的來信後，
24:18.452–24:20.000
建立了這個試算表。
24:20.000–24:21.479
接著，我們取得了這個輸出結果，
24:21.479–24:23.093
並將其轉換為一項技能。
24:23.093–24:24.902
我對這個輸出結果很滿意。
24:24.902–24:26.920
請將其轉換為一項技能。
24:26.920–24:29.947
現在，我可以重複執行這項操作，
24:29.947–24:32.469
無論是每週、每天或每小時。
24:32.469–24:40.575
我只需說：「請在每個星期五上午 9:00 執行此操作，
24:40.575–24:43.200
並更新這個試算表。」
24:43.200–24:46.269
這樣做會建立一個自動化流程
24:46.269–24:49.400
請注意，這裡我們只看到兩項自動化。
24:49.400–24:52.428
這裡顯示，「我將把它設定為每週五上午9點的定期自動化任務
24:52.428–24:54.360
每週五上午9點的定期自動化任務。」
24:54.360–24:56.040
我們可以直接在介面中看到。
24:56.040–24:58.806
已建立每週品牌合作提案表更新
24:58.806–25:02.711
如果我們點擊這裡，可以看到自動化任務的詳細資訊。
25:02.711–25:05.242
我們可以看到它在每週五上午9點。
25:05.242–25:07.022
我們可以看到它處於啟用狀態。
25:07.022–25:09.273
我們可以看到上次執行時間為空
25:09.273–25:11.214
因為它從未執行過
25:11.214–25:15.250
我們可以看到我們正在使用品牌合作提案研究員技能
25:15.250–25:20.762
來掃描 Gmail 中的付費推廣、贊助、品牌合作、影響者行銷活動等。
25:20.762–25:21.711
所以建立了這個
25:21.711–25:24.046
現在如果我們前往自動化任務標籤頁，
25:24.046–25:25.286
我們現在可以在這裡看到它。
25:25.286–25:26.835
我們可以看到每週五上午9點。
25:26.835–25:28.964
我們還有另一個在每週五上午9點。
25:28.964–25:30.087
我們可以點擊這裡
25:30.087–25:31.772
然後直接編輯它。
25:31.772–25:35.037
最後，我們在 Codex 內有一個額外功能
25:35.037–25:37.412
叫做紀事（Chronicle）。
25:37.412–25:39.723
所以這是 Codex 內一個全新的功能。
25:39.723–25:40.972
如果我們前往設定
25:40.972–25:42.377
然後點擊設定
25:42.377–25:44.251
然後前往個人化
25:44.251–25:45.500
然後往下捲動，
25:45.500–25:49.404
你會看到一個選項可以啟用 Chronicle 研究預覽。
25:49.404–25:50.660
當你開啟這個功能時，
25:50.660–25:53.381
這樣做會顯示它正在執行
25:53.381–25:55.892
這個功能看起來可能有點侵入性。
25:55.892–25:59.578
這個功能基本上會一直錄製你的螢幕
25:59.578–26:02.125
這樣它就能掌握你正在處理的內容。
26:02.125–26:04.142
所以如果我打開這裡的簡報
26:04.142–26:06.697
然後我在瀏覽器裡隨意捲動，
26:06.697–26:08.916
在瀏覽器裡隨意捲動，
26:08.916–26:09.800
使用 Chronicle。
26:09.800–26:18.440
呃，告訴我我應該在我的 Codex 視覺簡報中加入什麼。
26:18.440–26:22.951
然後它會顯示這裡使用 Chronicle 技能來處理即時螢幕工作內容
26:22.951–26:24.680
然後將所有內容交叉轉移。
26:24.680–26:26.074
Chronicle 可用
26:26.074–26:28.652
所以我先檢查即時螢幕狀態
26:28.652–26:29.628
看看這個，
26:29.628–26:33.680
你可以看到它正在擷取我螢幕上最近的截圖。
26:33.680–26:36.146
所以它正在抓取舊的圖片
26:36.146–26:41.967
並從我最近螢幕上開啟的內容中獲取上下文。
26:41.967–26:45.781
因此它能夠追蹤我正在處理的所有事項
26:45.781–26:49.040
而呼叫它的方式只是說「使用 Chronicle」。
26:49.040–26:50.691
所以它說基於 Chronicle，
26:50.691–26:52.814
我會將這些加入你的視覺簡報中。
26:52.814–26:54.383
一個 Codex 超級應用程式地圖，
26:54.383–26:57.697
一個用於提示檔案的完整控制迴路圖，
26:57.697–26:59.440
一個 Chronicle 示範投影片，
26:59.440–27:01.184
可重用的技能投影片，
27:01.184–27:03.887
以及為什麼選擇 Codex 而非其他工具的投影片。
27:03.887–27:05.271
所以，它自動地，
27:05.271–27:08.101
而我沒有提供任何[清喉嚨]資訊，
27:08.101–27:09.359
我沒有上傳圖片
27:09.359–27:11.183
或提供 Excalidraw 文件，
27:11.183–27:13.520
它就知道這份簡報投影片裡的內容。
27:13.520–27:15.458
而且它能夠理解它
27:15.458–27:18.715
因為這個 Chronicle 功能會監控我的螢幕
27:18.715–27:21.041
並持續擷取螢幕截圖
27:21.041–27:24.953
因此，這些是 Codex 內的主要功能與能力。
27:24.953–27:28.126
總結來說，我們介紹了 Codex 擁有完整的檔案控制權，
27:28.126–27:30.792
意味著你可以建立、編輯和刪除任何檔案。
27:30.792–27:32.117
它具備持久記憶功能，
27:32.117–27:36.030
無論是自動記憶還是你 agents.md 檔案中的手動記憶。
27:36.030–27:37.640
我們也討論了外掛程式，
27:37.640–27:39.507
這些外掛程式已內建於平台中。
27:39.507–27:42.587
它將你的代理程式與所有現有的工具連接起來。
27:42.587–27:47.074
你可以透過使用代理程式執行的任何任務
27:47.074–27:49.560
來建立技能，然後將其轉換為技能。
27:49.560–27:53.015
你的 Codex 代理程式可以存取影像生成，
27:53.015–27:56.808
並且它擁有全球最佳的影像生成模型，
27:56.808–27:59.589
也就是全球最佳的影像生成模型，
27:59.589–28:01.200
即 GPT-image-2。
28:01.200–28:03.796
我們還擁有瀏覽器控制與電腦使用功能，
28:03.796–28:07.400
這讓你的代理程式能夠像人類一樣控制你的電腦。
28:07.400–28:08.582
人類介面，
28:08.582–28:10.880
包括鍵盤、滑鼠和螢幕，
28:10.880–28:12.259
現在也可以由代理程式使用，
28:12.259–28:14.491
而且它正在呈指數級改善。
28:14.491–28:15.771
我們還有自動化功能。
28:15.771–28:18.006
你可以將代理執行過的任何有用任務
28:18.006–28:19.515
說：「嘿，每天做這個。
28:19.515–28:20.712
嘿，每小時做這個。
28:20.712–28:21.825
嘿，每週做這個。
28:21.825–28:23.618
每月做這個。」就是這類事情。
28:23.618–28:25.860
然後作為額外福利，我們談到了 Chronicle。
28:25.860–28:28.206
總之，非常感謝大家觀看。
28:28.206–28:30.035
我希望你們從這個影片中學到了很多。
28:30.035–28:32.074
我每週都會製作這類影片，
28:32.074–28:34.520
所以請務必訂閱並點擊讚按鈕。
28:34.520–28:35.310
這對我幫助很大。
28:35.310–28:38.840
我們下一個影片見。
0:00.000–0:02.852
You're in the right place if you use ChatGPT,
如果你使用 ChatGPT，
0:02.852–0:05.936
Claude, Claude Code, Claude Co-work, or Codex,
Claude、Claude Code、Claude Co-work 或 Codex，
0:05.936–0:09.800
but don't know which one you should use for business and how.
但不知道該在商業上如何使用它們，
0:09.800–0:13.840
Because in this video, we're talking about using AI agents effectively.
因為在這個影片中，我們將探討如何有效使用 AI 代理。
0:13.840–0:16.430
Specifically, we're talking about Codex,
因為在本影片中，我們將探討如何有效使用 AI 代理。
0:16.430–0:17.725
OpenAI's agent tool,
具體來說，我們將討論 Codex，
0:17.725–0:19.308
which is like Claude Code,
OpenAI 的代理工具，
0:19.308–0:24.129
except it's an easy-to-use interface for doing knowledge work and coding work.
它是一個易於使用的介面，用於處理知識工作和程式碼工作。
0:24.129–0:29.031
In this video, we're going to be breaking down the seven core capabilities of Codex,
但提供了一個易於使用的介面，用於進行知識工作和程式碼編寫。
0:29.031–0:30.711
OpenAI's AI agent super app,
在本影片中，我們將拆解 Codex 的七項核心功能，
0:30.711–0:32.602
each with real-world use cases,
OpenAI 的 AI 代理超級應用程式，
0:32.602–0:37.924
so that by the end, you know exactly [music] what it can do and where it fits in your workflow.
這樣到最後，你會清楚知道 [音樂] 它能做什麼，以及它如何融入你的工作流程。
0:37.924–0:39.853
Let's get it fits in your workflow.
讓你在看完後，能清楚知道 [音樂] 它能做什麼，以及它如何融入你的工作流程。
0:39.853–0:40.880
Let's get started.
讓它融入你的工作流程。
0:40.880–0:42.551
So, this right here is Codex.
所以，這裡就是 Codex。
0:42.551–0:47.961
This is a clean interface for AI agents that can fully control your computer.
這是一個乾淨的介面，讓 AI 代理能夠完全控制你的電腦。
0:47.961–0:51.738
At first glance, it kind of looks like ChatGPT or Claude,
這是一個乾淨的 AI 代理介面，可以完全控制你的電腦。
0:51.738–0:53.748
because you can create a chat,
乍看之下，它有點像 ChatGPT 或 Claude，
0:53.748–0:55.998
but they're not exactly the same.
因為你可以建立聊天，
0:55.998–0:59.271
When you chat on ChatGPT or Claude and upload files,
當你在 ChatGPT 或 Claude 上聊天並上傳檔案時，
0:59.271–1:02.164
those files are actually stored in the cloud.
這些檔案實際上儲存在雲端。
1:02.164–1:10.267
With Codex, all of the files that you upload to Codex or that you create with Codex are stored on your computer.
使用 Codex 時，你上傳到 Codex 或使用 Codex 建立的檔案都會儲存在你的電腦上。
1:10.267–1:14.880
Codex is made to do any task that you might want to do on a computer.
Codex 旨在執行你在電腦上可能想進行的任何任務。
1:14.880–1:17.480
You can create a motion graphics video.
你可以建立動態圖形影片。
1:17.480–1:19.666
You can create beautiful landing pages,
你可以建立精美的落地頁，
1:19.666–1:24.103
and directly inside Codex, you can generate images that you can place in your site.
並且直接在 Codex 內生成圖片，然後將其放置在你的網站上。
1:24.103–1:29.280
On Codex, you can also create more complex games and 3D simulations.
在 Codex 上，你也可以建立更複雜的遊戲和 3D 模擬。
1:29.280–1:34.464
You can also create mobile apps with a front end and back end with a professional design.
你也可以建立具有前端和後端且具備專業設計的行動應用程式。
1:34.464–1:38.056
You can also create desktop apps in just a few prompts.
你也可以僅透過幾個提示詞建立桌面應用程式。
1:38.056–1:42.554
You can ask GPT 5.5 to create you an Excel sheet after doing a ton of research,
你可以要求 GPT 5.5 在完成大量研究後為你建立 Excel 試算表，
1:42.554–1:45.767
and it can create spreadsheets with multiple slides.
並且它能建立包含多個工作表的試算表。
1:45.767–1:50.779
It can also create any type of Word doc with many different charts,
它也能建立包含多種不同圖表的任何類型 Word 文件，
1:50.779–1:59.254
and you can also create presentations directly inside Codex and export them to Canva or any other tool you want.
你也可以直接在 Codex 內建立簡報，並將其匯出到 Canva 或你想要的任何其他工具。
1:59.254–2:03.804
And I haven't even mentioned yet that Codex can fully control my computer,
我甚至還沒提到 Codex 可以完全控制我的電腦，
2:03.804–2:06.005
and it can also control the browser.
而且它也能控制瀏覽器。
2:06.005–2:08.248
But, this is something we'll talk about a little bit later.
但這是我們稍後會討論的內容。
2:08.248–2:12.315
As we go through the main seven different capabilities of Codex,
當我們逐步了解 Codex 的七項主要功能時，
2:12.315–2:14.951
you'll realize that it is truly a super app
你會發現它確實是一個超級應用程式，
2:14.951–2:16.533
that does any coding task
能夠執行任何程式碼任務
2:16.533–2:18.265
or any knowledge work task.
或任何知識性工作任務。
2:18.265–2:19.723
Let's not waste any time.
我們別浪費時間了。
2:19.723–2:22.012
Let's hop into capability number one.
讓我們直接進入第一個功能。
2:22.012–2:26.089
So, the first capability that Codex has is full file access.
所以，Codex 的第一個功能是擁有完整的檔案存取權限。
2:26.089–2:28.375
You see, when I use ChatGPT,
你看，當我使用 ChatGPT 時，
2:28.375–2:31.456
all of the files that I give to ChatGPT
我提供給 ChatGPT 的所有檔案
2:31.456–2:35.640
or that ChatGPT generates are stored in the cloud.
或是 ChatGPT 產生的檔案，都會儲存在雲端。
2:35.640–2:37.337
However, when we use Codex,
然而，當我們使用 Codex 時，
2:37.337–2:39.328
all of the files that are created
所有產生的檔案
2:39.328–2:42.795
or the files that I give to it are stored on your computer.
或是我提供給它的檔案，都會儲存在你的電腦上。
2:42.795–2:48.299
And the agent inside Codex has full access to your computer.
而且 Codex 內部的代理程式可以完全存取你的電腦。
2:48.299–2:49.339
Let me show you.
讓我展示給你看看。
2:49.339–2:51.150
In my downloads folder,
在我的下載資料夾中，
2:51.150–2:54.409
I have a folder called receipts to process.
我有一個名為「待處理收據」的資料夾。
2:54.409–2:57.386
And we just have 60 different little photos of receipts
裡面只有 60 張不同的收據小照片
2:57.386–2:59.069
that I've added to this folder.
是我新增到這個資料夾裡的。
2:59.069–3:00.317
I can simply ask Codex,
我只需要簡單地詢問 Codex，
3:00.317–3:01.762
"In the downloads folder,
「在下載資料夾裡，
3:01.762–3:04.258
you'll see my receipts that I need to process.
你會看到我需要處理的收據。
3:04.258–3:07.250
I want you to analyze them in an Excel sheet
我希望你將它們分析成 Excel 試算表
3:07.250–3:10.242
and create charts that helps me visualize
並建立圖表，幫助我視覺化
3:10.242–3:12.123
the types of transactions
交易類型
3:12.123–3:15.030
and any other patterns that you notice."
以及你注意到的任何其他模式。
3:15.030–3:16.996
So, we can choose our model,
所以，我們可以選擇我們的模型，
3:16.996–3:18.535
which is GPT 5.5 high,
也就是 GPT 5.5 high，
3:18.535–3:19.732
and we can run it.
然後執行它。
3:19.732–3:20.616
And the agent,
而這個代理程式，
3:20.616–3:22.606
because it has full file access,
因為它擁有完整的檔案存取權限，
3:22.606–3:24.891
it can access any file on my computer,
它可以存取我電腦上的任何檔案，
3:24.891–3:26.807
it'll find the relevant folder
它會找到相關的資料夾
3:26.807–3:28.207
and complete the task.
並完成任務。
3:28.207–3:29.359
And you can see here,
你可以在這裡看到，
3:29.359–3:30.918
it found the target folder,
它找到了目標資料夾，
3:30.918–3:34.171
which is users Riley Brown downloads receipts to process
也就是 users Riley Brown downloads receipts to process
3:34.171–3:35.459
with only 53 receipts.
裡面只有 53 張收據。
3:35.459–3:36.888
I'm going to OCR those,
我將對這些收據進行 OCR，
3:36.888–3:40.575
and so we're going to extract the text from those receipts,
因此我們將從這些收據中提取文字，
3:40.575–3:43.358
classify and categorize each transaction.
並對每筆交易進行分類和歸類。
3:43.358–3:47.470
Now, it's creating an Excel workbook on my computer
現在，它正在我的電腦上建立一個 Excel 工作簿
3:47.470–3:49.680
that we can open up in Codex.
我們可以在 Codex 中開啟它。
3:49.680–3:50.266
All right.
好的。
3:50.266–3:51.700
So, after 7 minutes,
所以，經過 7 分鐘後，
3:51.700–3:52.629
it is now done
現在已經完成了
3:52.629–3:56.341
and we can very easily just open this Excel project up
我們可以非常輕鬆地直接在
3:56.341–3:57.944
directly inside Code X
Code X 內部開啟這個 Excel 專案
3:57.944–4:01.066
and you can see that it has created this file.
你可以看到它已經建立了這個檔案。
4:01.066–4:02.839
We can see there's a dashboard,
我們可以看到有一個儀表板，
4:02.839–4:05.090
the total spend is there's a dashboard,
儀表板上顯示總支出，
4:05.090–4:06.600
the total spend is 25,982.
總支出為 25,982。
4:06.600–4:07.699
We see receipts,
我們可以看到收據，
4:07.699–4:08.955
category summary,
類別摘要，
4:08.955–4:10.133
payment summary,
付款摘要，
4:10.133–4:12.724
and we can see all the different trends,
以及我們可以看見所有不同的趨勢，
4:12.724–4:14.608
the spend by payment method.
按付款方式分列的支出。
4:14.608–4:16.344
We can see the monthly trends.
我們可以看到每月的趨勢。
4:16.344–4:17.809
This is an Excel sheet
這是一個 Excel 試算表
4:17.809–4:20.087
and because it's on our computer,
而且因為它位於我們的電腦上，
4:20.087–4:22.040
if you click this down arrow,
如果你點擊這個向下箭頭，
4:22.040–4:23.748
you can see open in folder
你可以看到「在資料夾中開啟」
4:23.748–4:28.142
and we can see the exact file location of this Excel spreadsheet.
我們可以看見這個 Excel 試算表的確切檔案位置。
4:28.142–4:32.870
So, now that we know that all of the files created on Code X live on your computer,
所以，現在我們知道在 Code X 上建立的所有檔案都保存在你的電腦上，
4:32.870–4:35.736
the question becomes how do you stay organized?
問題就變成瞭如何保持組織有序？
4:35.736–4:41.407
And I will tell you the best way that you can stay organized inside Code X is to work in a project.
我會告訴你，在 Code X 中保持組織有序的最佳方式就是在專案中工作。
4:41.407–4:44.280
So, you can create a chat in one of two ways.
因此，你可以透過兩種方式建立聊天。
4:44.280–4:50.558
One way you can create a chat is if you just go to this chats category and you just hit this new chat button.
建立聊天的一種方式是，如果你直接進入這個聊天類別並點擊這個新聊天按鈕。
4:50.558–4:54.674
It allows you to create a chat that just shows up here in the sidebar.
它允許你建立一個聊天，該聊天只會顯示在側邊欄中。
4:54.674–4:57.449
It's not actually inside a project.
它實際上並不在專案內部。
4:57.449–5:04.022
Or, you can go ahead and you can create a project and you can always hit start at existing folder.
或者，你可以繼續建立一個專案，並隨時點擊「開始於現有資料夾」。
5:04.022–5:11.012
all of my projects in my documents folder and then I will just create a new project seven capabilities
我的文件資料夾中的所有專案，然後我將建立一個新的專案七種能力
5:11.012–5:13.591
cuz that's what we're talking about.
因為我們正在討論的就是這個。
5:13.591–5:20.349
We can create a new folder and now we see that we're creating a new chat inside the seven capabilities folder.
我們可以建立一個新資料夾，現在我們看到我們正在七種能力資料夾內建立新聊天。
5:20.349–5:26.328
You'll see here this project will show up here in the left side panel and as soon as I add a chat,
你會看到這個專案會顯示在左側面板中，而一旦我新增聊天，
5:26.328–5:28.502
you can see the chat shows up here.
你就可以看到聊天顯示在這裡。
5:28.502–5:34.909
The cool part about Code X is you can multitask and you can create multiple chats at the same time.
Code X 的酷之處在於你可以多工處理，並同時建立多個聊天。
5:34.909–5:40.000
Here, I'm going to ask the agent to create a Word doc.
在這裡，我將要求代理建立一個 Word 文件。
5:40.000–5:44.790
And we're doing this within a chat inside the seven capabilities folder.
而我們是在七種能力資料夾內的聊天中進行此操作。
5:44.800–5:47.245
Any document that the agent creates
代理建立的任何文件
5:47.245–5:48.467
when it's in a chat
當它處於聊天中
5:48.467–5:49.608
inside a project
在專案內部
5:49.608–5:52.950
will actually store the document that it creates
實際上會儲存它所建立的文件
5:52.950–5:55.150
in this folder on your computer.
在你電腦上的這個資料夾中。
5:55.160–5:57.667
So, if we click these three dots here
所以，如果我們點擊這裡的三個點
5:57.667–5:59.172
and say open in finder
並選擇在訪客中開啟
5:59.172–6:02.013
and we open the seven categories folder,
然後開啟七個類別資料夾，
6:02.013–6:04.604
we'll see that there's nothing in it.
我們會發現裡面什麼都沒有。
6:04.604–6:07.638
And that's because the agent is not done creating the document.
這是因為代理程式尚未完成建立文件。
6:07.638–6:09.058
As soon as the agent's done,
一旦代理程式完成工作，
6:09.058–6:10.972
it's going to put as the agent's done,
它會顯示代理程式已完成，
6:10.972–6:14.306
it's going to put the uh document that it creates in this project.
它會將它在此專案中建立的文件放置於此。
6:14.306–6:16.139
And as you can see here,
正如你在這裡看到的，
6:16.139–6:19.130
we have this Word doc that was created
我們有一個已建立的 Word 文件
6:19.130–6:23.955
and it's called learn 90% of OpenAI Codex in under 2 minutes.
名為「兩分鐘內學會 OpenAI Codex 90% 的內容」。
6:23.955–6:26.602
You can see the title of the folder right here.
你可以直接在這裡看到資料夾的標題。
6:26.602–6:30.068
If we go ahead and open up finder at that location
如果我們繼續在該位置開啟訪客
6:30.068–6:32.408
and we click on outputs and docs,
並點擊輸出和文件，
6:32.408–6:36.221
you can see that that document is located right here.
你可以看到該文件位於這裡。
6:36.221–6:40.188
So, this is literally on my computer by default.
所以，這實際上預設就在我的電腦上。
6:40.188–6:45.122
And so, this video that I'm making now is called the seven capabilities of Codex.
因此，我現在製作的這部影片名為「Codex 的七項能力」。
6:45.122–6:48.852
And if we were to create a new chat in this folder,
如果我們要在這個資料夾中建立新的對話，
6:48.852–6:53.702
the chat would be aware that it is starting out in this project.
對話會知道它是在這個專案中開始的。
6:53.702–6:55.653
I can type at learn and see
我可以輸入 @learn 並查看
6:55.653–7:02.620
I can actually at mention any file that's created within this seven capabilities folder.
我實際上可以@提及在這個七項能力資料夾中建立的任何檔案。
7:02.620–7:04.483
And I can say,
我可以說，
7:04.483–7:09.902
"Please turn this into a landing page,
「請將這個轉成登入頁面，
7:09.902–7:15.840
a good learning resource for my viewers."
作為我觀眾的良好學習資源。」
7:15.840–7:16.186
And so,
因此，
7:16.186–7:18.203
I can take any document within this folder
我可以取用這個資料夾中的任何文件，
7:18.203–7:19.470
and I can just at mention it
然後直接@提及它，
7:19.470–7:19.874
and say,
並說，
7:19.874–7:21.660
"Please turn this to a landing page."
「請將這個轉成登入頁面。」
7:21.660–7:23.504
And because it can control my computer
因為它能控制我的電腦，
7:23.504–7:26.280
and go look at all of the documents that are organized very
並去查看所有組織得非常好的文件，
7:26.280–7:26.711
easily,
輕鬆地，
7:26.711–7:28.251
it can find the right document
它能找到正確的檔案，
7:28.251–7:29.853
and then create a landing page.
然後建立登入頁面。
7:29.853–7:33.332
And the code for that landing page will exist on my computer.
該登入頁面的程式碼將存在於我的電腦上。
7:33.332–7:34.840
And here we go, it's done.
好了，完成了。
7:34.840–7:41.919
We can open this up and we can see that it created a landing page based on that document that is stored on our computer.
我們可以打開它，看到它根據儲存在我們電腦上的那份文件建立了一個登入頁面。
7:41.919–7:44.729
And let's go ahead and move on to the next capability.
讓我們繼續進行下一個能力。
7:44.729–7:48.527
Next up, Codex has something called persistent memory.
接下來，Codex 擁有一項稱為持久記憶的功能。
7:48.527–7:50.812
So, within Codex, there's two types of memory.
所以，在 Codex 中有兩種類型的記憶。
7:50.812–7:52.505
There's manual memory,
有手動記憶，
7:52.505–7:54.283
and there's auto memory.
以及自動記憶。
7:54.283–7:57.752
Manual memory is when you ask the agent to remember something,
手動記憶是指你要求代理程式記住某些內容，
7:57.752–7:58.687
when you tell it,
當你告訴它，
7:58.687–8:01.889
"Hey, can you remember this going forward in the future?"
「嘿，你能記住這個，以便在未來使用嗎？」
8:01.889–8:03.357
Let me show you an example.
讓我舉個例子。
8:03.357–8:03.564
So,
所以，
8:03.564–8:05.838
I made a few changes to that landing page
我對那個登陸頁面做了一些修改，
8:05.838–8:07.905
and got it to create it in this format
並讓它以這種格式建立，
8:07.905–8:09.697
with this little left sidebar.
帶有這個小小的左側邊欄。
8:09.697–8:10.945
We can scroll down,
我們可以向下滾動，
8:10.945–8:14.688
and we can automatically snap to the section that we like.
並且可以自動跳轉到我們喜歡的區塊。
8:14.688–8:16.190
I really like this format,
我非常喜歡這種格式，
8:16.190–8:16.873
and we like.
我們喜歡。
8:16.873–8:18.886
I really like this format,
我非常喜歡這種格式，
8:18.886–8:20.259
and so from now on,
所以從現在開始，
8:20.259–8:25.017
can you please use this format whenever I ask for a landing page
能否請你在我要求建立登陸頁面時，
8:25.017–8:27.680
and I don't specify uh the styling?
且我沒有特別指定樣式時，
8:27.680–8:29.724
You should use this styling
使用這種樣式，
8:29.724–8:35.599
and remember that if there's a lot of text content to use this table of contents.
並記住，如果有大量文字內容，請使用這個目錄。
8:35.599–8:35.852
So,
所以，
8:35.852–8:39.230
I can simply just ask the agent to remember this,
我可以直接要求代理程式記住這個，
8:39.230–8:43.536
and it's automatically going to add it to the agents.md file,
它會自動將它新增到 agents.md 檔案中，
8:43.536–8:45.731
which is where you can see here,
你可以在這裡看到，
8:45.731–8:47.335
search for agents.md.
搜尋 agents.md。
8:47.335–8:50.865
This is where it's going to store this information for later.
這就是它將此資訊儲存以供日後使用的地方。
8:50.865–8:51.333
And so,
因此，
8:51.333–8:52.424
you can see here,
你可以在這裡看到，
8:52.424–8:54.840
the agent updated its manual memory,
代理程式更新了它的手動記憶體，
8:54.840–8:56.710
which is the agents.md file.
也就是 agents.md 檔案。
8:56.710–8:57.767
If we click on this,
如果我們點擊這裡，
8:57.767–9:00.806
here are the things that I've asked it to add in the past.
這裡是我過去要求它新增的內容。
9:00.806–9:02.773
Here's the what it just added.
這是它剛剛新增的內容。
9:02.773–9:06.658
It says for landing pages that are made for learning purposes,
它指出，對於用於學習目的的登陸頁面，
9:06.658–9:08.825
it says use this text first format,
它說使用這種文字優先格式，
9:08.825–9:12.112
um and then it kind of describes the style that I like.
嗯，然後它大致描述了我最喜歡的風格。
9:12.112–9:18.216
This agents.md file is a living document that gets updated over time
這個 agents.md 檔案是一個會隨時間更新的活文件，
9:18.216–9:21.000
that acts as the manual memory.
它充當手動記憶體的角色。
9:21.000–9:25.373
But, there's also a different type of memory that you should never
但是，還有另一種類型的記憶體是你應該永遠不要
9:25.373–9:30.859
and this is an auto memory feature that Codex kind of keeps updated automatically.
這是 Codex 會自動保持更新的自動記憶體功能。
9:30.859–9:33.197
And just like the agents.md file,
就像 agents.md 檔案一樣，
9:33.197–9:36.287
this is all stored in a file on your computer.
這些資料都儲存在你電腦上的某個檔案裡。
9:36.287–9:41.513
If we go to Codex and hit command N to create a new chat,
如果我們打開 Codex 並按下 Command N 來建立新的對話，
9:41.513–9:45.432
please tell me about what you have in the
請告訴我你在那個
9:45.432–9:48.757
and then Codex / memories folder.
在 Codex / memories 資料夾裡的內容。
9:48.757–9:53.516
I want to understand what you know.
我想了解你所知道的資訊。
9:53.516–9:56.031
Also, open that file.
同時，打開那個檔案。
9:56.031–10:00.106
And you can see here it gave me the path to this file.
你可以看到這裡顯示了該檔案的路徑。
10:00.106–10:02.525
We can go ahead and open this up.
我們可以繼續打開它。
10:02.525–10:05.214
And this is a document you should not touch.
這份文件你不應該去動它。
10:05.214–10:10.234
You can just kind of observe it to see what the agent observes about you.
你只需要觀察它，看看 AI 代理者對你觀察到了什麼。
10:10.234–10:13.720
And so every few weeks I'll actually go through this and take a look.
因此，我每隔幾週就會去檢查這份文件並查看內容。
10:13.720–10:19.380
And this actually gives me a nice little summary of all of the different tasks that I've asked my agent to do.
這份文件實際上為我要求代理者執行的所有不同任務提供了一個簡潔的摘要。
10:19.380–10:23.711
And it can be useful when helping other people build out similar workflows.
在協助其他人建立類似的工作流程時，這會很有用。
10:23.711–10:27.760
I do not recommend making manual changes to these memories, right?
我不建議手動修改這些記憶，對吧？
10:27.760–10:31.043
Just let the agent store the memories however it wants to
就讓代理者自行決定如何儲存記憶吧。
10:31.043–10:34.186
because the agent has a skill that keeps these updated
因為代理者具備讓這些記憶保持更新的技能。
10:34.186–10:36.072
and it will get better over time.
而且它會隨著時間推移變得更好。
10:36.072–10:39.457
The agents.md file on the other hand is your manual memory.
另一方面，agents.md 檔案是你的手動記憶。
10:39.457–10:41.520
This is meant for you to keep it updated.
這是要讓你來保持更新的。
10:41.520–10:43.540
You can tell the agent to update it
你可以指示代理程式更新它
10:43.540–10:45.993
or you can manually update it if you want.
或者如果你願意，也可以手動更新它。
10:45.993–10:50.720
The third huge capability that Codex has is plugins.
Codex 的第三項強大功能是外掛程式。
10:50.720–10:54.004
And so plugins are basically just Codex connections.
而外掛程式基本上就是 Codex 的連線。
10:54.004–11:02.856
They are installable, reusable bundles that connect the OpenAI Codex AI agent to external tools, apps, and workflows.
它們是可安裝、可重複使用的套件，用於將 OpenAI Codex AI 代理程式與外部工具、應用程式和工作流程相連接。
11:02.856–11:05.940
And this allows them to have other functionality
這使得它們能夠具備其他功能
11:05.940–11:09.174
and they can connect to all of the tools that you use.
並且可以連接到你使用的所有工具。
11:09.174–11:11.806
And so if you go to the plugins tab,
因此，如果你前往外掛程式標籤頁，
11:11.806–11:14.155
with two clicks I can add Slack
只需兩下點擊，我就能新增 Slack
11:14.155–11:15.659
and I can add Gmail.
以及新增 Gmail。
11:15.659–11:17.495
Once you've added Gmail,
一旦你新增了 Gmail，
11:17.495–11:22.216
and you can see here this plugin actually comes with some skills.
你可以在這裡看到這個外掛程式實際上附帶了一些技能。
11:22.216–11:23.866
So there's a Gmail skill
所以這裡有 Gmail 技能
11:23.866–11:27.497
and an inbox triage skill that is part of this plugin.
以及作為此外掛程式一部分的收件匣篩選技能。
11:27.497–11:28.053
Right now,
目前，
11:28.053–11:30.276
I think there's over 100 different plugins
我認為有超過 100 種不同的外掛程式
11:30.276–11:31.882
that allow you to connect Codex
可以讓你將 Codex
11:31.882–11:33.611
to the tools that you already use.
連接到你已經使用的工具。
11:33.611–11:33.814
So,
所以，
11:33.814–11:35.233
let's go over an example.
讓我們來看一個範例。
11:35.233–11:35.427
So,
所以，
11:35.427–11:36.722
I created a new project,
我建立了一個新專案，
11:36.722–11:39.507
which is a folder on my computer called brand deals,
也就是電腦上名為「品牌合作」的資料夾，
11:39.507–11:40.607
and I'm going to say,
然後我會說，
11:40.607–11:43.327
"I want you to look at my email from the past 2 weeks.
「我想請你查看我過去兩週的電子郵件。
11:43.327–11:46.271
Find all the brands offering to do paid promotion,
找出所有提供付費推廣的品牌，
11:46.271–11:47.813
do research on all of them,
對它們進行研究，
11:47.813–11:50.406
and put them in a table with relevant notes."
並將它們整理成包含相關備註的表格。」
11:50.406–11:52.158
And then I can type the at sign,
接著我可以輸入 @ 符號，
11:52.158–11:53.770
and here I can type in Gmail.
然後在這裡輸入 Gmail。
11:53.770–11:56.278
All of the plugins show up in this menu,
所有外掛都會顯示在這個選單中，
11:56.278–11:59.177
so we can very easily at mention the plugins.
因此我們可以非常輕鬆地以 @ 提及這些外掛。
11:59.177–12:02.799
And this is a quick way to see which plugins that you have access to.
這也是一種快速查看你可使用之外掛的方式。
12:02.799–12:05.483
And I really like using this menu,
我真的很喜歡使用這個選單，
12:05.483–12:08.560
and then we can go ahead and click send.
接著我們可以繼續點擊發送。
12:08.560–12:12.144
And now Codex will search through all of my recent emails
現在 Codex 會搜尋我最近的所有電子郵件，
12:12.144–12:14.965
because I've given it access via the plugin,
因為我已透過外掛授予它存取權限，
12:14.965–12:16.566
and it will get back to us.
它會回覆我們。
12:16.566–12:19.604
This is taking about like 5 minutes so far,
到目前為止大約花了 5 分鐘，
12:19.604–12:21.948
and I have another idea for a task
我還有另一個想讓 AI 代理執行的任務想法
12:21.948–12:23.597
I want an AI agent to do.
我希望由 AI 代理來執行。
12:23.597–12:23.798
So,
所以，
12:23.798–12:26.619
I can very easily just hit this new message button,
我可以非常輕鬆地按下這個新訊息按鈕，
12:26.619–12:28.701
and I can type out exactly what I want.
然後精確地輸入我想要的內容。
12:28.701–12:29.023
So,
所以，
12:29.023–12:33.440
I'm making my next video on CLIs and MCPs in Codex.
我正在製作關於 Codex 中 CLI 和 MCP 的下一部影片。
12:33.440–12:35.083
Please look at my Notion,
請查看我的 Notion，
12:35.083–12:36.804
and so I at mention Notion,
然後我在這裡提到 Notion，
12:36.804–12:37.977
which is a plugin,
這是一個外掛程式，
12:37.977–12:40.960
and find my high-quality long-form scripts.
並找到我高品質的長篇腳本。
12:40.960–12:43.845
Then I want you to please write a script on this topic.
然後我希望你針對這個主題撰寫腳本。
12:43.845–12:45.920
Use Excalidraw diagrams.
使用 Excalidraw 圖表。
12:45.920–12:46.907
This is a skill,
這是一項技能，
12:46.907–12:48.360
which I'll get to next.
我接下來會介紹。
12:48.360–12:49.657
We can run this prompt.
我們可以執行這個提示。
12:49.657–12:52.204
And so now we have two AI agents working,
現在我們有兩個 AI 代理正在運作，
12:52.204–12:53.980
both working with plugins.
兩者都使用外掛程式。
12:53.980–12:54.229
So,
所以，
12:54.229–12:56.225
the brand deals task is done.
品牌合作任務已完成。
12:56.225–12:59.676
If we go ahead and open up this document right here,
如果我們繼續打開這裡的文件，
12:59.676–13:01.812
we can see that it created file,
我們可以看到它建立了檔案，
13:01.812–13:04.195
and it went through all of my emails
並且它瀏覽了我所有的電子郵件
13:04.195–13:09.371
and wrote down the companies offering to sponsor my videos all the way down.
並詳細記錄了所有提供贊助我影片的公司資訊。
13:09.371–13:11.911
It literally went through all of them
它確實瀏覽了所有郵件
13:11.911–13:13.550
and took detailed notes
並做了詳細的筆記
13:13.550–13:14.205
and said,
並詢問，
13:14.205–13:15.844
"What did they ask for?"
「他們要求什麼？」
13:15.844–13:18.630
And also they added some research notes.
此外，它也加入了一些研究筆記。
13:18.630–13:21.467
And then I could ask it to respond to the emails
接著，我可以要求它回覆電子郵件
13:21.467–13:24.527
because of the same way that it can read my emails,
因為它能讀取我的電子郵件，
13:24.527–13:26.991
it can also send emails from my account.
它也能從我的帳號發送電子郵件。
13:26.991–13:29.751
And if we go to the Notion plugin that we tested,
如果我們前往之前測試過的 Notion 外掛，
13:29.751–13:32.440
we can see here that it created this document,
我們可以看到這裡它建立了這份文件，
13:32.440–13:34.209
and it went through my Notion,
並且它瀏覽了我的 Notion，
13:34.209–13:36.898
found all the scripts that I made previously,
找到了我之前製作的所有腳本，
13:36.898–13:37.606
and it said,
並說道，
13:37.606–13:41.710
"Do you ever feel like AI coding tools are getting more powerful every
「你是否覺得 AI 編碼工具每星期都變得更強大，
13:41.720–13:44.160
week, but also somehow more confusing?"
但同時卻也變得更令人困惑？」
13:44.160–13:46.055
And this is literally in my voice
這確實是用我的語氣
13:46.055–13:48.510
because it went through all of my scripts,
因為它瀏覽了我所有的腳本，
13:48.510–13:51.668
and it created this document using the Notion plugin.
並使用 Notion 外掛程式建立了這份文件。
13:51.668–13:55.082
And that brings us to the fourth capability that Codex has,
這就帶我們來到 Codex 的第四項功能，
13:55.082–13:56.127
which are skills.
也就是技能（Skills）。
13:56.127–14:02.003
And you can think of skills on Codex as reusable workflow recipes or SOPs
你可以將 Codex 中的技能視為可重複使用的工作流程配方或標準作業程序（SOP），
14:02.003–14:06.410
that your agent can use many times over and over again.
讓你的代理程式可以反覆多次使用。
14:06.410–14:09.018
In Codex, if you go up to the plugins tab,
在 Codex 中，如果你點擊上方的外掛程式（Plugins）標籤，
14:09.018–14:12.653
you see that there's actually a subtab within plugins,
你會發現外掛程式標籤下其實有一個子標籤，
14:12.653–14:13.838
which are skills.
那就是技能（Skills）。
14:13.838–14:18.025
And in the skills tab, these are similar to plugins.
在技能標籤中，這些功能與外掛程式類似。
14:18.025–14:20.042
They're more like instruction files
它們更像是指示檔案，
14:20.042–14:24.012
that your agent will follow every time it wants to use one of these skills.
當你的代理程式想要使用其中一項技能時，它會遵循這些指示。
14:24.012–14:27.095
So, all your skills can be found in the plugins folder
因此，你可以在外掛程式資料夾
14:27.095–14:28.566
in the skills subfolder.
的技能子資料夾中找到所有技能。
14:28.566–14:31.190
But, how do you actually create a skill?
但是，你實際上該如何建立技能呢？
14:31.190–14:34.049
In order to create a skill, you can type a prompt just like this.
為了建立技能，你可以輸入這樣的提示詞。
14:34.049–14:39.171
I want you to create a skill called brand deal analysis that does blank.
我希望你建立一個名為「品牌合作分析」的技能，功能為空白。
14:39.171–14:42.611
And I could enter this prompt in and create this skill,
我可以輸入這個提示詞並建立這個技能，
14:42.611–14:44.293
and then once that's done,
然後一旦完成，
14:44.293–14:47.580
in order to use any skill, you just press this slash.
要使用任何技能，你只需按下這個斜線鍵。
14:47.580–14:53.390
And so, I can use my Excalidraw diagram skill just by use my Excalidraw diagram skill just by typing slash.
因此，我只需輸入斜線鍵，就能使用我的 Excalidraw 圖表技能，只要輸入斜線鍵即可使用我的 Excalidraw 圖表技能。
14:53.390–14:55.573
I can also use my YouTube researcher skill
我也可以使用我的 YouTube 研究員技能
14:55.573–14:57.195
or my YouTube thumbnail skill,
或是我的 YouTube 縮圖技能，
14:57.195–14:58.941
which we'll get to in just a skill,
我們馬上就會談到這個技能，
14:58.941–15:00.760
which we'll get to in just a second.
我們馬上就會談到。
15:00.760–15:04.382
These are all skills that I can quickly access by typing slash.
這些都是我可以透過輸入斜線快速存取的技能。
15:04.382–15:06.567
These are different than plugins,
這些與外掛程式不同，
15:06.567–15:08.526
which I can access by typing at.
外掛程式我可以透過輸入 at 符號來存取。
15:08.526–15:12.600
So, there's two main ways to create a skill in Codex.
因此，在 Codex 中建立技能主要有兩種方式。
15:12.600–15:14.361
One is the prompt to skill,
一種是提示轉技能，
15:14.361–15:19.726
which is basically just I want you to create a skill called blank that does blank.
基本上就是「我想要你建立一個名為空白、執行空白功能的技能」。
15:19.726–15:22.878
This will give you a reusable instructions file,
這會給你一個可重複使用的指示檔案，
15:22.878–15:23.955
which is a skill,
也就是技能，
15:23.955–15:27.040
but I don't think it will be as good as option one.
但我認為它不會比選項一好。
15:27.040–15:29.633
The best way to create a skill, in my opinion,
我認為建立技能的最佳方式，
15:29.633–15:31.806
is to use the manual workflow method,
是使用手動工作流程方法，
15:31.806–15:34.820
where you basically ask your agent to do something,
基本上就是要求你的代理程式執行某項任務，
15:34.820–15:35.521
it'll do it,
它會執行，
15:35.521–15:38.325
and then you basically go back and forth, right?
然後你基本上會來回調整，對吧？
15:38.325–15:39.798
Tell it to create something,
告訴它建立某樣東西，
15:39.798–15:41.761
and say, "Hey, I want you to change it."
然後說：「嘿，我希望你修改它。」
15:41.761–15:43.602
And you say, "Get the output right."
然後你說：「把輸出結果弄對。」
15:43.602–15:45.565
If you're not happy, you iterate more,
如果你不滿意，就繼續迭代，
15:45.565–15:47.836
and you kind of go back through this process,
然後你大致上會重新走過這個流程，
15:47.836–15:50.658
and you loop around until you're happy with the output.
並循環這個過程，直到你對輸出結果滿意為止。
15:50.658–15:52.843
And then you reverse engineer the skill,
接著你逆向工程這項技能，
15:52.843–15:55.671
and you will basically end up saying something like,
你基本上最終會說出類似下面的話：
15:55.671–15:57.214
"I'm happy with this output.
「我對這個輸出結果很滿意。
15:57.214–15:58.502
Turn it into a skill."
把它變成一項技能。」
15:58.502–16:00.147
And I'll actually show you.
我實際上會展示給你。
16:00.147–16:01.784
If we go back to Codex,
如果我們回到 Codex，
16:01.784–16:03.329
and you can see here,
你可以在這裡看到，
16:03.329–16:09.330
I've asked the agent to put all those paid promos into a spreadsheet right here.
我已經要求代理程式將所有付費促銷內容放入這裡的試算表中。
16:09.330–16:11.508
And I really like this spreadsheet.
我非常喜歡這個試算表。
16:11.508–16:12.880
I think it's high quality.
我認為它的品質很高。
16:12.880–16:15.023
I think it's color-coded, right,
我認為它已經依不同優先級進行了色彩編碼，對吧？
16:15.023–16:16.784
by the different priority.
依不同的優先級。
16:16.784–16:18.367
This is looking really good.
這看起來真的很不錯。
16:18.367–16:19.098
And remember,
並且記住，
16:19.098–16:20.559
you can always full screen it
你隨時可以將其全螢幕顯示
16:20.559–16:24.040
to get a better look at whatever document it is that you're creating.
以便更清楚地查看你正在建立的任何文件。
16:24.040–16:24.428
And so,
所以，
16:24.428–16:25.720
well, all we need to do is
我們只需要
16:25.720–16:27.206
we need to just type this in.
輸入這些內容。
16:27.206–16:28.600
"I'm happy with this output.
「我對這個輸出結果很滿意。
16:28.600–16:30.936
Turn this into a skill that I can use."
把它變成一項我可以使用的技能。」
16:30.936–16:33.422
And the agent is smart enough to be like,
而這個 AI 代理足夠聰明，會理解說，
16:33.422–16:38.772
"Okay, this is he wants to always be able to analyze his Gmail for brand sponsorships.
「好的，他總是希望能夠分析他的 Gmail 來尋找品牌贊助。
16:38.772–16:39.961
Alternate into a skill."
將其轉換為一項技能。」
16:39.961–16:40.301
And so,
所以，
16:40.301–16:40.980
now it's done.
現在已經完成了。
16:40.980–16:41.572
It says,
它顯示，
16:41.572–16:47.920
"I've created and validated the reusable skill called brand deal researcher skill.md."
「我已建立並驗證了一個名為 brand deal researcher skill.md 的可重用技能。」
16:47.920–16:50.286
And we can open up this skill by uh
我們可以透過點擊
16:50.286–16:52.564
clicking on the skill.md file,
skill.md 檔案來開啟這個技能，
16:52.564–16:55.105
and we can actually read this over.
並且我們實際上可以閱讀這些內容。
16:55.105–16:59.231
And it turned our entire workflow into these instructions.
它將我們整個工作流程轉化為這些指示。
16:59.231–17:03.618
And the AI is very smart enough at taking anything that you've done with your agent,
而 AI 非常聰明，能夠將你與代理互動的任何內容，
17:03.618–17:05.081
and turning it into a skill.
轉化為一項技能。
17:05.081–17:05.370
So,
所以，
17:05.370–17:08.738
now if we hit command N to create a new chat,
現在如果我們按下 Command N 來建立新的對話，
17:08.738–17:11.432
"I want you uh to please do {slash}
「我希望你呃，請執行 {slash}
17:11.432–17:14.223
It's called uh to please do {slash}
這叫做呃，請執行 {slash}
17:14.223–17:17.120
It's called brand deal researcher.
這叫做品牌合作研究員。
17:17.120–17:17.509
Right?
對吧？
17:17.509–17:17.754
So,
所以，
17:17.754–17:21.748
now we have this reusable skill that we can use at any point.
現在我們擁有一個可重複使用的技能，可以在任何時候使用。
17:21.748–17:22.761
And you can see here
你可以在這裡看到
17:22.761–17:24.596
it's using all of the same plugins.
它使用了所有相同的插件。
17:24.596–17:26.710
It's using the Gmail plugin.
它使用了 Gmail 插件。
17:26.710–17:26.955
So,
所以，
17:26.955–17:30.878
a skill can contain instructions to use a specific plugin
技能可以包含使用特定插件的指令
17:30.878–17:32.757
because at the end of the day
因為归根结底
17:32.757–17:35.699
it's just instructions to do a simple task.
它只是執行簡單任務的指令。
17:35.699–17:39.085
While our research brand deal task is loading,
當我們的品牌合作研究任務載入時，
17:39.085–17:45.423
I actually went ahead and did a create Excalidraw graphics for the seven capabilities.
我實際上已經先行建立了七項能力的 Excalidraw 圖形。
17:45.423–17:46.810
The video that we're doing now,
我們現在正在製作的影片，
17:46.810–17:48.943
remember this document that we created earlier
記得我們之前建立的這份文件嗎
17:48.943–17:53.797
which is this landing page that describes all of the different things that we're talking about in this video.
也就是這份介紹了本影片中所討論各種不同內容的登陸頁面。
17:53.797–17:55.582
I went ahead and I just said,
我已經先行說了，
17:55.582–18:00.937
"Please look at that document and make a ton of graphics on a single Excalidraw doc.
「請查看那份文件，並在單一 Excalidraw 文件中製作大量圖形。
18:00.937–18:05.755
Use the Excalidraw diagram skill and make the chart super useful."
使用 Excalidraw 圖表技能，製作一張極具實用價值的圖表。
18:05.755–18:11.960
So, then the AI thought and worked for 4 minutes and 23 seconds and then it created this.
接著，AI 思考並運作了 4 分 23 秒，然後產生了這個結果。
18:11.960–18:13.816
And we can open this up and we can just
我們可以打開它，然後直接
18:13.816–18:19.261
And we can open this up and we can just automatically basically import this project and boom, look at this.
我們可以打開它，然後基本上自動匯入這個專案，哇，看看這個。
18:19.261–18:23.360
It created this entire outline and I love this formatting.
它建立了整個大綱，我喜歡這種格式。
18:23.360–18:27.320
This formatting that it did with the background here, I actually really like that.
它在這裡使用背景所做的格式，我其實非常喜歡。
18:27.320–18:30.357
And so, it created all of the diagrams.
因此，它建立了所有的圖表。
18:30.357–18:32.204
Wow, this is actually fantastic.
哇，這真的非常棒。
18:32.204–18:33.672
It has the intro,
它有介紹，
18:33.672–18:35.559
all of the summaries,
所有的摘要，
18:35.559–18:36.293
memory,
記憶體，
18:36.293–18:38.390
plugins versus skills,
外掛程式與技能的比較，
18:38.390–18:39.858
skill creation.
技能建立。
18:39.858–18:40.944
Very cool.
非常酷。
18:40.944–18:42.765
So, now I can go back to Codex
所以，現在我可以回到 Codex
18:42.765–18:47.121
and I can say, "Wow, I really like the formatting that you did here.
然後我說：「哇，我真的很喜歡你這裡的格式。
18:47.121–18:48.550
Please update the skills
請更新技能
18:48.550–18:51.748
so that you always put them in the containers that you did
以便你總是將它們放在你所做的容器中
18:51.748–18:53.790
and I want you to always give me a link
我希望你總是給我一個連結
18:53.790–18:56.036
with all of the diagrams that you create
包含你所建立的所有圖表
18:56.036–18:58.417
when I ask you to create multiple diagrams
當我要求你建立多個圖表時
18:58.417–18:59.302
just like this.
就像這樣。
18:59.302–19:00.910
I love it."
我非常喜歡。
19:00.920–19:02.769
And so, even once you've used a skill,
因此，即使你已經使用過某項技能，
19:02.769–19:04.319
maybe you've used it for a while
也許你已經使用它一段時間了
19:04.319–19:05.035
and you like it
而且你喜歡它
19:05.035–19:08.911
and then you get surprised by something valuable like this diagram right here
然後你對像這裡這樣的圖表所帶來的價值感到驚訝
19:08.911–19:10.223
or these this formatting,
或是這種格式設定，
19:10.223–19:12.320
you can always just say, "Hey, I want you to
你總是可以說：「嘿，我希望你
19:12.320–19:13.998
always format it like this."
總是這樣格式化它。」
19:13.998–19:18.821
And so, every time you use the skill is an opportunity to make the skill even better.
因此，每次使用這項技能都是一次讓技能變得更好的機會。
19:18.821–19:21.944
Just tell your agent to update the skill to always do something.
只需告訴你的代理程式更新技能，以始終執行某項操作。
19:21.944–19:27.356
The next capability of Codex is built-in GPT image to access.
Codex 的下一項功能是內建的 GPT 圖像生成能力。
19:27.356–19:33.038
So, you can generate images with the best image model in the world directly inside Codex.
因此，你可以在 Codex 內部直接使用全球最佳的圖像模型來生成圖片。
19:33.038–19:35.994
So, I created a new project called content,
因此，我建立了一個名為 content 的新專案，
19:35.994–19:41.003
and here I'm saying, "Generate a product photo for my uh sweater company.
這裡我說：「為我的毛衣公司生成一張產品照片。
19:41.003–19:43.296
This is our first sweater.
這是我們的第一件毛衣。
19:43.296–19:45.199
I can paste an image of a sweater,
我可以貼上一張毛衣的圖片，
19:45.199–19:46.398
and you can see here,
你可以在這裡看到，
19:46.398–19:47.948
this is the first sweater,
這是第一件毛衣，
19:47.948–19:49.922
knitted sweater that I'm making.
我正在製作的針織毛衣。
19:49.922–19:53.506
Please create five images with GPT/image/2
請使用 GPT/image/2 建立五張圖片
19:53.506–19:55.153
of different models
由不同模特兒
19:55.153–19:57.865
of different of different models
不同模型的技能
19:57.865–20:00.190
of different nationalities
來自不同國籍
20:00.190–20:02.030
wearing this sweater.
穿著這件毛衣。
20:02.030–20:05.587
Three of the images should have one person,
其中三張圖片應各有一人，
20:05.587–20:08.551
one image should have three people,
一張圖片應有三個人，
20:08.551–20:11.920
the last image should have five people."
最後一張圖片應有五個人。”
20:11.920–20:13.411
And so, we can run this,
所以，我們可以執行這個，
20:13.411–20:15.451
and within the content folder,
並在 content 資料夾中，
20:15.451–20:17.334
it's going to create images.
它會建立圖片。
20:17.334–20:20.064
And those images will live inside the content folder
一旦完成，這些圖片將存放在 content 資料夾內
20:20.064–20:20.870
once it's done.
完成後。
20:20.870–20:21.653
And look at this.
看看這個。
20:21.653–20:26.136
You can see here that it's using the skill.md for image gen.
你可以在這裡看到它正在使用 skill.md 進行圖片生成。
20:26.136–20:29.500
And so, this image gen skill is built into Codex.
因此，這個圖片生成技能已內建於 Codex 中。
20:29.500–20:32.052
And so, you can go to Codex, click on skills,
所以，你可以前往 Codex，點擊技能，
20:32.052–20:35.242
and you can see which skills are built into the system.
並查看系統內建了哪些技能。
20:35.242–20:37.375
And you don't have to enable them,
你不需要啟用它們，
20:37.375–20:40.118
but these are recommended skills by Codex,
但這些是 Codex 推薦的技能，
20:40.118–20:41.880
and one of them is image gen.
其中一項是圖像生成。
20:41.880–20:45.218
And by default, this uses the latest and best image model.
預設情況下，這會使用最新且最佳的圖像模型。
20:45.218–20:48.481
And you can see here that it generated one of the images.
你可以在這裡看到它生成了一張圖片。
20:48.481–20:50.320
We still need to wait for the other four.
我們仍需等待其餘四張圖片。
20:50.320–20:52.827
Okay, here, it just generated the second image.
好的，這裡它剛剛生成了第二張圖片。
20:52.827–20:54.727
It just generated the third image.
它剛剛生成了第三張圖片。
20:54.727–20:57.823
The next image should have three people in one image.
下一張圖片應該會在一張圖片中包含三個人。
20:57.823–20:58.840
And here we go.
好了，我們繼續。
20:58.840–21:00.179
Here is number four,
這是第四張，
21:00.179–21:01.754
and here is number five.
這是第五張。
21:01.754–21:04.871
And now that you know that you can generate images,
現在你已經知道可以生成圖片了，
21:04.871–21:07.764
that brings me to the next capability of Codex,
這就引出了 Codex 的下一個功能，
21:07.764–21:10.139
which is browser use and computer use.
也就是瀏覽器使用和電腦使用。
21:10.139–21:13.396
And so in order to use computer use, it is a plugin.
因此，要使用電腦使用功能，它是一個外掛程式。
21:13.396–21:16.273
So remember, you just type in at computer use.
所以請記住，你只需輸入 @computer use。
21:16.273–21:20.079
Now I can say, "Please open the Canva app on my computer,
現在我可以說：「請在我的電腦上開啟 Canva 應用程式，
21:20.079–21:21.816
make a new presentation,
建立一份新的簡報，
21:21.816–21:25.539
and put each one of these images on each of the slides."
並將這些圖片中的每一張放到每一張投影片上。」
21:25.539–21:27.111
So one image per slide,
所以每張投影片放一張圖片
21:27.111–21:29.510
there should be five total slides.
總共應該有五張投影片。
21:29.520–21:32.085
We can simply tell Codex to control our computer
我們可以簡單地告訴 Codex 來控制我們的電腦
21:32.085–21:32.983
and it will do it.
它就會照做。
21:32.983–21:34.686
And as you can see here,
正如你在這裡看到的，
21:34.686–21:36.120
we see list Mac apps
我們看到列出的 Mac 應用程式
21:36.120–21:38.361
and now it's looking at Canva.
現在它正在查看 Canva。
21:38.361–21:39.502
You can see here,
你可以在這裡看到，
21:39.502–21:41.622
it's controlling my computer.
它正在控制我的電腦。
21:41.622–21:42.931
I'm not even touching it.
我甚至都沒有碰它。
21:42.931–21:45.006
It just created this presentation.
它剛剛建立了這份簡報。
21:45.006–21:46.746
It's clicking in Canva.
它正在 Canva 裡點擊。
21:46.746–21:47.234
Okay.
好的。
21:47.234–21:49.444
Oh, it just ran a script,
噢，它剛剛執行了一個腳本，
21:49.444–21:50.438
I believe.
我相信是這樣。
21:50.438–21:52.400
And so it just ran a script.
所以它只是執行了一個腳本。
21:52.400–21:54.303
It's looking at Canva
它正在查看 Canva
21:54.303–21:55.994
and you can see here
你可以在這裡看到
21:55.994–21:58.636
it created this presentation
它建立了這份簡報
21:58.636–22:02.336
and it has all five images on the computer.
並且電腦裡有所有五張圖片。
22:02.336–22:03.565
And you can see here,
你可以在這裡看到，
22:03.565–22:04.649
this is the mouse.
這是滑鼠。
22:04.649–22:08.302
You can see the mouse outline of the computer use.
你可以看到電腦使用的滑鼠游標。
22:08.302–22:11.081
It will move around from time to time.
它會不時地移動
22:11.081–22:14.360
It created this presentation for me.
它為我建立了這份簡報。
22:14.360–22:16.305
So this right here is a browser
所以這裡是一個瀏覽器。
22:16.305–22:17.549
and you can see that
你可以看到
22:17.549–22:19.572
if we literally exit out of this
如果我們直接退出這個
22:19.572–22:20.738
and hit plus sign,
並點擊加號，
22:20.738–22:21.983
we can see browser.
我們可以看到瀏覽器。
22:21.983–22:24.461
And when we open this index.html file,
當我們開啟這個 index.html 檔案時，
22:24.461–22:27.171
it's literally opening this in a browser.
它實際上是在瀏覽器中開啟這個檔案。
22:27.171–22:29.027
So now I'm going to say,
所以現在我要說，
22:29.027–22:30.882
"Turn this into an app,
「將這個轉換成應用程式，
22:30.882–22:32.836
then test the interface
然後測試介面
22:32.836–22:37.914
and make sure the buttons and navigation work using at browser
並確保按鈕和導航功能在瀏覽器中使用正常
22:37.914–22:41.440
and navigation work using at browser use."
且導航功能在瀏覽器使用中運作正常。」
22:41.440–22:43.161
And I can run this prompt here.
我可以執行這裡的提示。
22:43.161–22:45.060
And now it's going to take this,
現在它將處理這個，
22:45.060–22:46.229
turn it into an app,
將其轉換成應用程式，
22:46.229–22:48.274
and then it's just going to test it
然後它只會測試它
22:48.274–22:50.840
by controlling the browser inside Codex.
透過在 Codex 內部控制瀏覽器。
22:50.840–22:52.004
Okay, you can see here,
好的，你可以在這裡看到，
22:52.004–22:53.842
it's kind of like a different color.
它的顏色有點不一樣。
22:53.842–22:55.985
Look, this is my mouse right here.
看，這就是我的滑鼠指標。
22:55.985–22:57.596
This is the browser use.
這就是瀏覽器使用功能。
22:57.596–23:00.295
So the AI is actually controlling the browser
所以 AI 實際上正在控制瀏覽器
23:00.295–23:02.496
and it's going to start testing them.
並且它將開始進行測試。
23:02.496–23:05.285
So, we see that it's test start button,
所以，我們看到它點擊了測試開始按鈕，
23:05.285–23:08.596
and the AI is now taking control of the screen.
而 AI 現在正在掌控螢幕。
23:08.596–23:11.450
The first visible check showed the app controls,
第一個可見的檢查顯示了應用程式的控制項，
23:11.450–23:15.626
but the start button smooth scroll is too subtle for live verification.
但開始按鈕的平滑滾動對於即時驗證來說太微妙了。
23:15.626–23:17.425
You can see here it's moving around.
你可以在這裡看到它正在移動。
23:17.425–23:19.720
It's scrolling through the app.
它正在瀏覽應用程式。
23:19.720–23:27.120
It's clicking on the different quizzes that it's created.
它正在點擊它所建立的不同測驗。
23:27.120–23:29.088
Now, it's testing quiz completion,
現在，它正在測試測驗完成情況，
23:29.088–23:31.515
and now it just marked this a correct answer.
然後它剛才標記了這個正確答案。
23:31.515–23:41.080
So, it's making sure that all parts of the application work.
所以，它正在確保應用程式的所有部分都能運作。
23:41.080–23:42.975
Oh, now it tested the side panel,
哦，現在它測試了側邊面板，
23:42.975–23:44.600
and now we're at the bottom.
而我們現在已經到達底部。
23:44.600–23:45.592
This is very cool.
這非常酷。
23:45.592–23:48.725
Anything you can open up in the browser,
任何你可以在瀏覽器中開啟的內容，
23:48.725–23:50.623
you can ask browser use.
都可以使用瀏覽器使用者功能。
23:50.623–23:54.595
You can use the @browseruse plugin
你可以使用 @browseruse 外掛程式
23:54.595–23:58.440
to test it directly inside Codex.
直接在 Codex 內部進行測試。
23:58.440–24:02.846
And finally, capability number seven inside Codex is automation.
最後，Codex 內的第七項功能是自動化。
24:02.846–24:06.407
You can create automation simply by talking to Codex.
你可以透過與 Codex 對話來簡單地建立自動化。
24:06.407–24:09.868
So, remember this Excel sheet that we created in this chat here
所以，請記住我們在這裡的對話中建立的這個 Excel 試算表，
24:09.868–24:13.595
in the brand deals project called research paid promo brand deals.
在名為 research paid promo brand deals 的品牌合作專案中。
24:13.595–24:18.452
We created this spreadsheet after it analyzed all of my incoming emails
我們在使用電子郵件外掛程式分析了我所有的來信後，
24:18.452–24:20.000
with the email plugin.
建立了這個試算表。
24:20.000–24:21.479
Then, we took this output,
接著，我們取得了這個輸出結果，
24:21.479–24:23.093
and we turned it into a skill.
並將其轉換為一項技能。
24:23.093–24:24.902
I'm happy with this output.
我對這個輸出結果很滿意。
24:24.902–24:26.920
Please turn it into a skill.
請將其轉換為一項技能。
24:26.920–24:29.947
Now, what I can do is I can repeatedly do this
現在，我可以重複執行這項操作，
24:29.947–24:32.469
on a weekly or daily or hourly basis.
無論是每週、每天或每小時。
24:32.469–24:40.575
I can simply say, "Please do this every single Friday at 9:00 a.m., please,
我只需說：「請在每個星期五上午 9:00 執行此操作，
24:40.575–24:43.200
and update this sheet."
並更新這個試算表。」
24:43.200–24:46.269
What this is going to do is it's going to create an automation.
這樣做會建立一個自動化流程
24:46.269–24:49.400
Notice here we only see two automations.
請注意，這裡我們只看到兩項自動化。
24:49.400–24:52.428
Here it says, "I'll set this up as a recurring automation for Fridays at 9:00
這裡顯示，「我將把它設定為每週五上午9點的定期自動化任務
24:52.428–24:54.360
recurring automation for Fridays at 9:00 a.m."
每週五上午9點的定期自動化任務。」
24:54.360–24:56.040
And we can see it directly in the UI.
我們可以直接在介面中看到。
24:56.040–24:58.806
Created weekly brand deal sheet updated
已建立每週品牌合作提案表更新
24:58.806–25:02.711
and if we click on this, we can see the automation details.
如果我們點擊這裡，可以看到自動化任務的詳細資訊。
25:02.711–25:05.242
We can see that at Friday at 9:00 a.m.
我們可以看到它在每週五上午9點。
25:05.242–25:07.022
We can see that it's active.
我們可以看到它處於啟用狀態。
25:07.022–25:09.273
We can see that the last run is empty
我們可以看到上次執行時間為空
25:09.273–25:11.214
because it's never run before
因為它從未執行過
25:11.214–25:15.250
and we can see that we are using the brand deal researcher skill
我們可以看到我們正在使用品牌合作提案研究員技能
25:15.250–25:20.762
to scan Gmail for paid promotion sponsorship brand deal influencer campaigns, etc.
來掃描 Gmail 中的付費推廣、贊助、品牌合作、影響者行銷活動等。
25:20.762–25:21.711
So created this
所以建立了這個
25:21.711–25:24.046
and now if we go to the automations tab,
現在如果我們前往自動化任務標籤頁，
25:24.046–25:25.286
we now see it up here.
我們現在可以在這裡看到它。
25:25.286–25:26.835
We can see Friday at 9:00 a.m.
我們可以看到每週五上午9點。
25:26.835–25:28.964
We have another one at Fridays at 9:00 a.m.
我們還有另一個在每週五上午9點。
25:28.964–25:30.087
We can click on this
我們可以點擊這裡
25:30.087–25:31.772
and you can edit it directly.
然後直接編輯它。
25:31.772–25:35.037
And finally, we have a bonus capability inside Codex
最後，我們在 Codex 內有一個額外功能
25:35.037–25:37.412
which is something called Chronicle.
叫做紀事（Chronicle）。
25:37.412–25:39.723
So this is a brand new feature within Codex.
所以這是 Codex 內一個全新的功能。
25:39.723–25:40.972
If we go to settings
如果我們前往設定
25:40.972–25:42.377
and we click settings
然後點擊設定
25:42.377–25:44.251
and we go to personalization
然後前往個人化
25:44.251–25:45.500
and we scroll down,
然後往下捲動，
25:45.500–25:49.404
you will see an option to to use Chronicle research preview.
你會看到一個選項可以啟用 Chronicle 研究預覽。
25:49.404–25:50.660
When you turn this on,
當你開啟這個功能時，
25:50.660–25:53.381
what this will do is it'll say that it is running
這樣做會顯示它正在執行
25:53.381–25:55.892
and this one may seem a little bit invasive.
這個功能看起來可能有點侵入性。
25:55.892–25:59.578
What this one does is it basically is always recording your screen
這個功能基本上會一直錄製你的螢幕
25:59.578–26:02.125
so it has context with what you're working on.
這樣它就能掌握你正在處理的內容。
26:02.125–26:04.142
And so if I go to my presentation here
所以如果我打開這裡的簡報
26:04.142–26:06.697
and I just kind of scroll around in my browser,
然後我在瀏覽器裡隨意捲動，
26:06.697–26:08.916
use kind of scroll around in my browser,
在瀏覽器裡隨意捲動，
26:08.916–26:09.800
use Chronicle.
使用 Chronicle。
26:09.800–26:18.440
Uh tell me what I should add to my Codex visual presentation.
呃，告訴我我應該在我的 Codex 視覺簡報中加入什麼。
26:18.440–26:22.951
And so it shows you here all use Chronicle skill for live screen work content
然後它會顯示這裡使用 Chronicle 技能來處理即時螢幕工作內容
26:22.951–26:24.680
then cross everything over.
然後將所有內容交叉轉移。
26:24.680–26:26.074
Chronicle's available
Chronicle 可用
26:26.074–26:28.652
so I'm checking the fresh screen state first
所以我先檢查即時螢幕狀態
26:28.652–26:29.628
and look at this,
看看這個，
26:29.628–26:33.680
you can see that it's pulling up recent screenshots of my screen here.
你可以看到它正在擷取我螢幕上最近的截圖。
26:33.680–26:36.146
And so it is pulling old images
所以它正在抓取舊的圖片
26:36.146–26:41.967
and it's getting context from what I've had open on my screen recently.
並從我最近螢幕上開啟的內容中獲取上下文。
26:41.967–26:45.781
And so it's able to keep track of all the things that I'm working on
因此它能夠追蹤我正在處理的所有事項
26:45.781–26:49.040
and the way that you invoke it is just say use Chronicle.
而呼叫它的方式只是說「使用 Chronicle」。
26:49.040–26:50.691
And so it says based on Chronicle,
所以它說基於 Chronicle，
26:50.691–26:52.814
I'd add these to your visual presentation.
我會將這些加入你的視覺簡報中。
26:52.814–26:54.383
A Codex super app map,
一個 Codex 超級應用程式地圖，
26:54.383–26:57.697
a full control loop diagram for prompt files,
一個用於提示檔案的完整控制迴路圖，
26:57.697–26:59.440
a Chronicle demo slide,
一個 Chronicle 示範投影片，
26:59.440–27:01.184
reusable skills slide,
可重用的技能投影片，
27:01.184–27:03.887
and why Codex over other tools slide.
以及為什麼選擇 Codex 而非其他工具的投影片。
27:03.887–27:05.271
And so, it automatically,
所以，它自動地，
27:05.271–27:08.101
and I didn't give it any [clears throat] information,
而我沒有提供任何[清喉嚨]資訊，
27:08.101–27:09.359
I didn't upload an image
我沒有上傳圖片
27:09.359–27:11.183
or give it an Excalidraw document,
或提供 Excalidraw 文件，
27:11.183–27:13.520
it just knew what was in this slide deck here.
它就知道這份簡報投影片裡的內容。
27:13.520–27:15.458
And it's able to understand it
而且它能夠理解它
27:15.458–27:18.715
because this Chronicle feature watches my screen
因為這個 Chronicle 功能會監控我的螢幕
27:18.715–27:21.041
and takes screenshots constantly.
並持續擷取螢幕截圖
27:21.041–27:24.953
And so, those are the main features and abilities within Codex.
因此，這些是 Codex 內的主要功能與能力。
27:24.953–27:28.126
And in summary, we covered that Codex has full file control,
總結來說，我們介紹了 Codex 擁有完整的檔案控制權，
27:28.126–27:30.792
meaning you can create, edit, and delete any file.
意味著你可以建立、編輯和刪除任何檔案。
27:30.792–27:32.117
It has persistent memory,
它具備持久記憶功能，
27:32.117–27:36.030
and whether that is automatic memory or manual memory in your agents.md file.
無論是自動記憶還是你 agents.md 檔案中的手動記憶。
27:36.030–27:37.640
We also talked about plugins,
我們也討論了外掛程式，
27:37.640–27:39.507
which are built into the platform.
這些外掛程式已內建於平台中。
27:39.507–27:42.587
It connects your agent to all of the tools that exist.
它將你的代理程式與所有現有的工具連接起來。
27:42.587–27:47.074
You can create skills by taking any task that you create with the agent
你可以透過使用代理程式執行的任何任務
27:47.074–27:49.560
and then converting that into a skill.
來建立技能，然後將其轉換為技能。
27:49.560–27:53.015
Your Codex agent has access to image generation,
你的 Codex 代理程式可以存取影像生成，
27:53.015–27:56.808
and it is the best image generation model in the world,
並且它擁有全球最佳的影像生成模型，
27:56.808–27:59.589
which is generation model in the world,
也就是全球最佳的影像生成模型，
27:59.589–28:01.200
which is GPT-image-2.
即 GPT-image-2。
28:01.200–28:03.796
We also have browser control and computer use,
我們還擁有瀏覽器控制與電腦使用功能，
28:03.796–28:07.400
and this allows your agent to control your computer like a human.
這讓你的代理程式能夠像人類一樣控制你的電腦。
28:07.400–28:08.582
The human interface,
人類介面，
28:08.582–28:10.880
which is a keyboard, a mouse, and a screen,
包括鍵盤、滑鼠和螢幕，
28:10.880–28:12.259
can now be used by agents,
現在也可以由代理程式使用，
28:12.259–28:14.491
and it's getting better exponentially.
而且它正在呈指數級改善。
28:14.491–28:15.771
And we also have automations.
我們還有自動化功能。
28:15.771–28:18.006
You can take any useful task that your agent does
你可以將代理執行過的任何有用任務
28:18.006–28:19.515
and say, "Hey, do this every day.
說：「嘿，每天做這個。
28:19.515–28:20.712
Hey, do this every hour.
嘿，每小時做這個。
28:20.712–28:21.825
Hey, do this every week.
嘿，每週做這個。
28:21.825–28:23.618
Do this every month." That type of thing.
每月做這個。」就是這類事情。
28:23.618–28:25.860
And then as a bonus, we talked about Chronicle.
然後作為額外福利，我們談到了 Chronicle。
28:25.860–28:28.206
Anyway, thank you guys so much for watching.
總之，非常感謝大家觀看。
28:28.206–28:30.035
I hope you got a lot out of this video.
我希望你們從這個影片中學到了很多。
28:30.035–28:32.074
I make videos like this every single week,
我每週都會製作這類影片，
28:32.074–28:34.520
so make sure to subscribe and hit that like button.
所以請務必訂閱並點擊讚按鈕。
28:34.520–28:35.310
It helps me out a lot.
這對我幫助很大。
28:35.310–28:38.840
I'll see you here for the next video.
我們下一個影片見。
影片筆記：Learn 95% of Codex in 30 minutes
一句話總結
OpenAI 推出的 AI Agent 工具 Codex 是一個具備「完全控制電腦」能力的超級應用程式，透過完整檔案存取、持久記憶、外掛連接、技能創建、影像生成、瀏覽器/電腦控制以及自動化功能，實現從知識工作到程式開發的全流程自動化，且所有產出預設儲存於使用者本地電腦。
核心重點
- 本地化與完全控制：與 ChatGPT 或 Claude 不同，Codex 生成的檔案預設儲存於使用者本地電腦，並具備對電腦和瀏覽器的完全控制權。
- 三大核心能力：
- 完整檔案存取：支援專案管理，自動將產出物儲存至指定資料夾，支援多工處理與檔案引用。
- 持久記憶：包含透過
agents.md的手動記憶，以及由 Agent 自動維護的自動記憶。 - 外掛程式 (Plugins)：連接 Gmail、Slack、Notion 等外部工具，擴展工作流。
- 技能 (Skills) 與自動化：可將反覆執行或滿意的操作流程轉化為可重複使用的「技能」，並設定為定期執行的「自動化」任務。
- 內建生成與控制能力：內建 GPT-image/2 進行圖像生成，並透過
@computer use和@browser use插件控制桌面應用與瀏覽器。 - Chronicle 新功能：透過持續記錄螢幕畫面獲取上下文，讓 AI 能根據使用者當前操作自動建議內容。
詳細大綱
A. Codex 基本介紹與定位
- 目標受眾：使用 ChatGPT、Claude、Claude Code、Claude Co-work 或 Codex，但不確定如何將其用於商業知識工作與程式開發的使用者。
- 核心定義：OpenAI 的 AI Agent 超級應用程式 (Super App)，提供易於使用的介面，用於知識工作 (Knowledge Work) 與程式開發 (Coding Work)。
- 與競品差異：
- ChatGPT/Claude：上傳的檔案儲存於雲端。
- Codex：上傳或生成的檔案全部儲存於使用者本地電腦。
- 功能範圍：
- 生成動態圖形影片 (Motion graphics video)。
- 建立登陸頁面 (Landing pages)。
- 生成圖片並放置於網站。
- 建立複雜遊戲與 3D 模擬。
- 建立具備前後端與專業設計的行動應用程式 (Mobile apps)。
- 建立桌面應用程式 (Desktop apps)。
- 生成 Excel 試算表、Word 文件、簡報 (可匯出至 Canva)。
- 完全控制電腦與瀏覽器。
B. 核心能力一：完整檔案存取 (Full File Access)
- 檔案儲存機制：
- Agent 擁有電腦的完整檔案存取權限。
- 所有生成的檔案預設儲存於本地電腦，而非雲端。
- 專案管理 (Project Organization)：
- 建立聊天 (Chat)：僅顯示在側邊欄，未關聯到特定資料夾。
- 建立專案 (Project)：
- 可從現有資料夾開始 (Start at existing folder)。
- 專案會顯示在左側面板。
- 在專案內的聊天所生成的文件，會自動儲存至該專案對應的本地資料夾中。
- 多工處理 (Multitasking)：
- 可同時建立多個聊天視窗。
- 支援「@提及」(At mention) 專案內的檔案，引用現有文件進行後續操作（例如將 Word 文件轉為登陸頁面）。
- 實戰案例：發票處理：
- 情境：處理資料夾內的 60 張發票照片。
- 操作：指定模型為 GPT 5.5 high。
- 流程：Agent 識別資料夾 -> OCR 提取文字 -> 分類交易 -> 建立 Excel 工作簿。
- 結果：生成包含儀表板、總支出、類別摘要、付款摘要及趨勢圖表的 Excel 檔案，儲存於本地。
C. 核心能力二：持久記憶 (Persistent Memory)
Codex 具備兩種記憶類型，均儲存於本地檔案中：
- 手動記憶 (Manual Memory)：
- 機制：使用者要求 Agent 記住特定偏好或格式。
- 儲存位置：
agents.md檔案。 - 特性：是一個「活的文件」(Living document)，會隨時間更新。使用者可手動更新或要求 Agent 更新。
- 案例：使用者要求 Agent 記住特定的登陸頁面樣式（左側欄、目錄格式），Agent 會將此偏好寫入
agents.md。
- 自動記憶 (Auto Memory)：
- 機制：Agent 自動觀察並更新使用者行為與任務歷史。
- 儲存位置：
Codex / memories資料夾內的檔案。 - 特性：建議使用者不要手動修改，讓 Agent 自行維護以確保準確性。
- 用途：提供使用者任務歷史摘要，可用於協助他人建立類似工作流。
D. 核心能力三：外掛程式 (Plugins)
- 定義：可安裝、可重複使用的套件，用於連接 OpenAI Codex 與外部工具、應用程式及工作流。
- 功能：擴展 Agent 功能，連接使用者日常使用的工具。
- 操作方式：
- 透過 Plugins 標籤頁安裝（例如：Slack、Gmail）。
- 安裝後，外掛可能包含特定技能 (Skills)，如 Gmail 技能、收件匣分類技能 (Inbox triage skill)。
- 目前支援超過 100 種外掛。
- 引用方式：在聊天中使用「@提及」(At mention) 外掛名稱來調用。
- 實戰案例：品牌合作與腳本撰寫：
- 案例 1 (Gmail)：
- 情境：尋找過去兩週的品牌推廣郵件。
- 操作：建立專案資料夾，要求 Agent 搜尋 Gmail，研究品牌並建立包含相關註解的表格。
- 引用：使用
@Gmail外掛。 - 案例 2 (Notion)：
- 情境：撰寫關於 CLIs 和 MCPs 的影片腳本。
- 操作：要求 Agent 查看 Notion 中的高質量長篇腳本，並使用 Excalidraw 圖表撰寫新腳本。
- 引用：使用
@Notion外掛。
E. 技能 (Skills) 的應用與創建
- 技能的概念
- 技能是可重複使用的工作流程配方 (Workflow recipes) 或標準作業程序 (SOPs)。
- 技能類似於插件，但本質上是指示文件 (Instruction files)，代理 (Agent) 每次使用時都會遵循。
- 位置：位於插件 (Plugins) 資料夾下的技能 (Skills) 子資料夾。
- 調用方式：輸入斜線 (
/) 即可快速訪問技能；插件則使用@符號。
- 創建技能的方法
- 方法一：提示詞轉技能 (Prompt to Skill)
- 格式：「我想要你創建一個名為 [名稱] 的技能，功能是 [功能]」。
- 特點：生成可重複使用的指示文件，但講者認為效果可能不如方法二。
- 方法二：手動工作流程法 (Manual Workflow Method) - 推薦
- 流程：要求代理執行任務 -> 反覆調整與迭代直到滿意 -> 反向工程 (Reverse engineer) 將流程轉化為技能。
- 關鍵指令：「我對這個輸出很滿意，將其轉化為我可以使用的技能 (Turn this into a skill that I can use)」。
- 結果：代理會生成
.md檔案（如brand deal researcher skill.md），將整個工作流程轉化為指示。
- 技能的更新與優化
- 技能可以包含調用特定插件的指示。
- 使用技能後若發現更好的格式或功能，可要求代理更新技能，使其未來總是採用該格式。
- 每次使用技能都是優化技能的機會。
F. 內建圖像生成 (Built-in GPT Image)
- 功能說明
- Codex 內建 GPT-image/2 模型，可直接在系統內生成圖片。
- 屬於系統內建的推薦技能 (Recommended skills)，預設使用最新且最佳的圖像模型，無需手動啟用。
- 操作示範
- 創建新專案 (Content)。
- 上傳參考圖片（如毛衣）。
- 提示詞要求：生成不同國家模特兒穿著該毛衣的五張圖片（分別為 1 人、1 人、1 人、3 人、5 人的組合）。
- 結果：圖片生成後保存在內容資料夾中，並使用內建的
image gen skill.md。
G. 瀏覽器與電腦控制 (Browser Use & Computer Use)
- 電腦控制 (Computer Use)
- 插件調用：
@computer use。 - 功能：控制桌面應用程式。
- 示範：打開 Canva 應用程式，創建簡報，並將生成的圖片逐一放入幻燈片中。
- 特徵：使用者可見滑鼠游標移動，代理自動點擊與操作，無需使用者介入。
- 瀏覽器控制 (Browser Use)
- 插件調用：
@browser use。 - 功能：控制瀏覽器進行測試與互動。
- 示範：將 HTML 檔案轉為應用程式，並測試介面按鈕與導航功能。
- 特徵：代理控制瀏覽器游標，執行點擊、滾動、驗證按鈕功能等測試步驟。
H. 自動化 (Automation)
- 功能說明
- 透過與 Codex 對話即可創建自動化任務。
- 可將已創建的技能設定為定期執行（每週、每日或每小時）。
- 操作示範
- 基於之前創建的「品牌合作研究技能 (Brand deal researcher skill)」。
- 指令：「請在每個星期五上午 9:00 執行此操作，並更新此試算表」。
- 結果：
- UI 中顯示自動化任務狀態（Active）。
- 設定為每週五上午 9:00 重複執行。
- 使用
brand deal researcher skill掃描 Gmail 以獲取品牌贊助資訊。
I. 新功能：Chronicle 研究預覽版
- 啟用方式：Settings > Personalization > 開啟「Chronicle research preview」。
- 運作機制：
- 持續記錄螢幕畫面。
- 獲取使用者當前工作內容的上下文。
- 看似具有侵入性（invasive），但能提供即時情境感知。
- 操作示範：
- 指令：「use Chronicle」。
- 請求：「告訴我應該在我的 Codex 視覺簡報中添加什麼」。
- 結果：AI 識別近期螢幕截圖，建議添加「Codex super app map」、「prompt files 的全控制迴圈圖表」、「Chronicle 演示投影片」等內容。
- 關鍵點：無需上傳圖片或 Excalidraw 文件，AI 僅憑螢幕監控即可理解簡報內容。
J. Codex 功能總覽回顧
- 檔案控制 (Full file control)：建立、編輯、刪除任何檔案。
- 記憶系統 (Persistent memory)：
- 自動記憶。
- 手動記憶（透過
agents.md檔案）。 - 外掛 (Plugins)：平台內建，連接 Agent 至所有工具。
- 技能 (Skills)：將 Agent 執行的任務轉換為技能。
- 影像生成 (Image generation)：
- 支援影像生成。
- 宣稱為全球最佳模型：GPT-image-2。
- 瀏覽器控制與電腦使用 (Browser control and computer use)：
- Agent 可像人類一樣控制電腦。
- 介面包含鍵盤、滑鼠、螢幕。
- 效能持續指數級提升。
- 自動化 (Automations)：
- 將任務設定為定期執行（每日、每小時、每週、每月）。
工具 / 模型 / 名詞整理
- OpenAI Codex (或稱 Code X)：影片主要介紹的 AI Agent 工具。
- ChatGPT：對比對象，檔案儲存於雲端。
- Claude：對比對象，檔案儲存於雲端。
- Claude Code：對比對象。
- Claude Co-work：對比對象。
- GPT 5.5 high：影片中選擇使用的模型版本。
- GPT 5.5：提及的模型名稱。
- GPT-image/2 (或 GPT-image-2)：Codex 內建使用的圖像生成模型名稱。
- Canva：桌面應用程式名稱，用於創建簡報；簡報匯出目標工具。
- Slack：Codex 支援連接的外掛工具。
- Gmail：Codex 支援連接的外掛工具。
- Notion：Codex 支援連接的外掛工具。
- Excalidraw：用於繪製圖表的工具；提及「give it an Excalidraw document」。
- Finder：macOS 檔案管理系統（提及「Open in Finder」）。
- agents.md：儲存手動記憶的檔案名稱。
- Codex / memories：儲存自動記憶的資料夾路徑。
- @computer use：控制電腦應用程式的插件調用方式。
- @browser use：控制瀏覽器的插件調用方式。
- @Gmail：引用 Gmail 外掛的方式。
- @Notion：引用 Notion 外掛的方式。
- brand deal researcher skill.md：講者手動創建的技能檔案名稱。
- image gen skill.md：Codex 內建的圖像生成技能檔案。
- Excalidraw diagram skill：用於創建圖表的技能。
- YouTube researcher skill：用於研究 YouTube 相關內容的技能。
- YouTube thumbnail skill：用於處理 YouTube 縮圖的技能。
- Chronicle：Codex 內的新功能/模組名稱，位於 Settings > Personalization。
- index.html：瀏覽器中打開的檔案類型。
- CLIs and MCPs：提及影片主題包含 "CLIs" 與 "MCPs"。
- seven capabilities folder：使用者建立專案資料夾時命名。
- receipts to process：資料夾名稱。
- learn 90% of OpenAI Codex in under 2 minutes：生成的 Word 文件標題。
操作流程整理
流程一：建立專案與處理發票 (完整檔案存取)
- 建立專案 (Project)，選擇從現有資料夾開始。
- 指定模型為 GPT 5.5 high。
- Agent 識別資料夾內的發票照片。
- Agent 執行 OCR 提取文字並分類交易。
- Agent 建立 Excel 工作簿，包含儀表板、總支出、類別摘要、付款摘要及趨勢圖表。
- 檔案自動儲存至專案對應的本地資料夾。
流程二：創建與使用技能 (Skills)
- 創建技能 (手動工作流程法)：
- 要求 Agent 執行特定任務。
- 反覆調整與迭代直到滿意。
- 輸入指令：「我對這個輸出很滿意，將其轉化為我可以使用的技能」。
- Agent 生成
.md檔案 (如brand deal researcher skill.md)。
- 使用技能：
- 在聊天中輸入斜線 (
/) 訪問技能列表。 - 選擇並執行該技能。
- 更新技能：
- 若發現更好格式，要求 Agent 更新技能，使其未來採用該格式。
流程三：圖像生成 (內建 GPT Image)
- 創建新專案 (Content)。
- 上傳參考圖片 (如毛衣)。
- 輸入提示詞要求生成不同國家模特兒穿著該毛衣的圖片。
- Agent 使用內建
image gen skill.md生成圖片。 - 圖片保存至內容資料夾。
流程四：電腦與瀏覽器控制
- 電腦控制：
- 調用
@computer use插件。 - Agent 自動打開桌面應用 (如 Canva)。
- Agent 自動點擊、操作，將生成的圖片放入幻燈片。
- 瀏覽器控制：
- 調用
@browser use插件。 - Agent 控制瀏覽器游標，執行點擊、滾動、驗證按鈕功能等測試步驟。
流程五：設定自動化 (Automation)
- 基於已創建的技能 (如
brand deal researcher skill)。 - 輸入指令：
生字列表
| 生字 | 讀音 | 類型 | 中文 |
|---|---|---|---|
| agent | /ˈeɪdʒənt/ | 代理程式 | |
| interface | /ˈɪntərfeɪs/ | noun | 介面 |
| workflow | /ˈwɜːrkfloʊ/ | noun | 工作流程 |
| capability | /ˌkeɪpəˈbɪləti/ | noun | 功能；能力 |
| landing page | /ˈlændɪŋ peɪdʒ/ | noun | 落地頁 |
| prompt | /prɑːmpt/ | noun | 提示詞 |
| spreadsheet | /ˈspredʃiːt/ | noun | 試算表 |
| dashboard | /ˈdæʃbɔːrd/ | noun | 儀表板 |
| organize | /ˈɔːrɡənaɪz/ | verb | 組織；整理 |
| multitask | /ˌmʌltiˈtæsk/ | verb | 多工處理 |
| persistent | /pərˈsɪstənt/ | adjective | 持久的；持續的 |
| plugin | /ˈplʌɡɪn/ | noun | 外掛程式 |
| iterate | /ˈɪtəreɪt/ | verb | 迭代；反覆修改 |
| automation | /ˌɔːtəˈmeɪʃn/ | noun | 自動化 |
| invasive | /ɪnˈveɪsɪv/ | adjective | 侵入性的 |
| context | /ˈkɑːntekst/ | noun | 上下文；背景資訊 |
| invoke | /ɪnˈvoʊk/ | verb | 呼叫；調用 |
| exponentially | /ˌekspəˈnenʃəli/ | adverb | 指數級地 |
生字解說
agent /ˈeɪdʒənt/
意思：代理程式
解說：在 AI 語境中，指能自主執行任務、控制工具或處理資訊的軟體實體。
影片原句
Specifically, we're talking about Codex, OpenAI's agent tool.
具體來說，我們將討論 Codex，OpenAI 的代理工具。
Specifically, we're talking about Codex, OpenAI's agent tool.
具體來說，我們將討論 Codex，OpenAI 的代理工具。
延伸例句
The AI agent can browse the web and summarize articles.
該 AI 代理程式可以瀏覽網頁並摘要文章。
The AI agent can browse the web and summarize articles.
該 AI 代理程式可以瀏覽網頁並摘要文章。
interface /ˈɪntərfeɪs/
意思：介面
解說：使用者與系統互動的畫面或操作層，此處指 Codex 提供給使用者的操作視窗。
影片原句
except it's an easy-to-use interface for doing knowledge work and coding work.
但提供了一個易於使用的介面，用於進行知識工作和程式碼編寫。
except it's an easy-to-use interface for doing knowledge work and coding work.
但提供了一個易於使用的介面，用於進行知識工作和程式碼編寫。
延伸例句
The new user interface makes the software easier to navigate.
新的使用者介面讓軟體更易于導航。
The new user interface makes the software easier to navigate.
新的使用者介面讓軟體更易于導航。
workflow /ˈwɜːrkfloʊ/
意思：工作流程
解說：完成特定任務所需的一系列步驟或程序。
影片原句
so that by the end, you know exactly what it can do and where it fits in your workflow.
這樣到最後，你會清楚知道它能做什麼，以及它如何融入你的工作流程。
so that by the end, you know exactly what it can do and where it fits in your workflow.
這樣到最後，你會清楚知道它能做什麼，以及它如何融入你的工作流程。
延伸例句
This plugin integrates seamlessly into your existing workflow.
這個外掛程式能無縫整合到你現有的工作流程中。
This plugin integrates seamlessly into your existing workflow.
這個外掛程式能無縫整合到你現有的工作流程中。
capability /ˌkeɪpəˈbɪləti/
意思：功能；能力
解說：指系統或工具所具備的特定執行任務的本領或特性。
影片原句
In this video, we're going to be breaking down the seven core capabilities of Codex.
在本影片中，我們將拆解 Codex 的七項核心功能。
In this video, we're going to be breaking down the seven core capabilities of Codex.
在本影片中，我們將拆解 Codex 的七項核心功能。
延伸例句
The camera has advanced capabilities like night mode and zoom.
這款相機具有夜間模式和變焦等先進功能。
The camera has advanced capabilities like night mode and zoom.
這款相機具有夜間模式和變焦等先進功能。
landing page /ˈlændɪŋ peɪdʒ/
意思：落地頁
解說：網路行銷中，使用者點擊廣告後到達的第一個網頁，通常用於轉換或資訊傳遞。
影片原句
You can create beautiful landing pages, and directly inside Codex, you can generate images that you can place in your site.
你可以建立精美的落地頁，並且直接在 Codex 內生成圖片，然後將其放置在你的網站上。
You can create beautiful landing pages, and directly inside Codex, you can generate images that you can place in your site.
你可以建立精美的落地頁，並且直接在 Codex 內生成圖片，然後將其放置在你的網站上。
延伸例句
We optimized the landing page to increase conversion rates.
我們優化了落地頁以提高轉換率。
We optimized the landing page to increase conversion rates.
我們優化了落地頁以提高轉換率。
prompt /prɑːmpt/
意思：提示詞
解說：使用者輸入給 AI 的指令或文字，用以引導 AI 產生特定輸出。
影片原句
You can also create desktop apps in just a few prompts.
你也可以僅透過幾個提示詞建立桌面應用程式。
You can also create desktop apps in just a few prompts.
你也可以僅透過幾個提示詞建立桌面應用程式。
延伸例句
A clear prompt helps the AI generate more accurate results.
清晰的提示詞有助於 AI 產生更準確的結果。
A clear prompt helps the AI generate more accurate results.
清晰的提示詞有助於 AI 產生更準確的結果。
spreadsheet /ˈspredʃiːt/
意思：試算表
解說：用於儲存、整理和分析數據的電子表格文件，如 Excel。
影片原句
It can also create spreadsheets with multiple slides.
並且它能建立包含多個工作表的試算表。
It can also create spreadsheets with multiple slides.
並且它能建立包含多個工作表的試算表。
延伸例句
Please update the spreadsheet with the latest sales figures.
請用最新的銷售數據更新試算表。
Please update the spreadsheet with the latest sales figures.
請用最新的銷售數據更新試算表。
dashboard /ˈdæʃbɔːrd/
意思：儀表板
解說：整合顯示關鍵數據和指標的視覺化介面。
影片原句
We can see there's a dashboard, the total spend is there's a dashboard, the total spend is 25,982.
我們可以看到有一個儀表板，儀表板上顯示總支出，總支出為 25,982。
We can see there's a dashboard, the total spend is there's a dashboard, the total spend is 25,982.
我們可以看到有一個儀表板，儀表板上顯示總支出，總支出為 25,982。
延伸例句
The marketing dashboard shows real-time campaign performance.
行銷儀表板顯示了活動的即時表現。
The marketing dashboard shows real-time campaign performance.
行銷儀表板顯示了活動的即時表現。
organize /ˈɔːrɡənaɪz/
意思：組織；整理
解說：將事物安排得井井有條，此處指管理檔案和專案結構。
影片原句
the question becomes how do you stay organized?
問題就變成瞭如何保持組織有序？
the question becomes how do you stay organized?
問題就變成瞭如何保持組織有序？
延伸例句
It is important to organize your files into folders.
將檔案整理到資料夾中很重要。
It is important to organize your files into folders.
將檔案整理到資料夾中很重要。
multitask /ˌmʌltiˈtæsk/
意思：多工處理
解說：同時進行多項任務或操作。
影片原句
The cool part about Code X is you can multitask and you can create multiple chats at the same time.
Code X 的酷之處在於你可以多工處理，並同時建立多個聊天。
The cool part about Code X is you can multitask and you can create multiple chats at the same time.
Code X 的酷之處在於你可以多工處理，並同時建立多個聊天。
延伸例句
Modern operating systems allow users to multitask efficiently.
現代作業系統允許使用者有效率地進行多工處理。
Modern operating systems allow users to multitask efficiently.
現代作業系統允許使用者有效率地進行多工處理。
persistent /pərˈsɪstənt/
意思：持久的；持續的
解說：指記憶或狀態能夠長期保留，不會因對話結束而消失。
影片原句
Next up, Codex has something called persistent memory.
接下來，Codex 擁有一項稱為持久記憶的功能。
Next up, Codex has something called persistent memory.
接下來，Codex 擁有一項稱為持久記憶的功能。
延伸例句
The company has a persistent commitment to sustainability.
該公司對永續發展有持續的承諾。
The company has a persistent commitment to sustainability.
該公司對永續發展有持續的承諾。
plugin /ˈplʌɡɪn/
意思：外掛程式
解說：附加到主程式中以擴展其功能的軟體組件。
影片原句
The third huge capability that Codex has is plugins.
Codex 的第三項強大功能是外掛程式。
The third huge capability that Codex has is plugins.
Codex 的第三項強大功能是外掛程式。
延伸例句
You can install a plugin to add password protection.
你可以安裝外掛程式來增加密碼保護功能。
You can install a plugin to add password protection.
你可以安裝外掛程式來增加密碼保護功能。
iterate /ˈɪtəreɪt/
意思：迭代；反覆修改
解說：重複執行某個過程以逐步改進結果。
影片原句
If you're not happy, you iterate more, and you kind of go back through this process.
如果你不滿意，就繼續迭代，然後你大致上會重新走過這個流程。
If you're not happy, you iterate more, and you kind of go back through this process.
如果你不滿意，就繼續迭代，然後你大致上會重新走過這個流程。
延伸例句
We need to iterate on the design based on user feedback.
我們需要根據使用者回饋對設計進行迭代。
We need to iterate on the design based on user feedback.
我們需要根據使用者回饋對設計進行迭代。
automation /ˌɔːtəˈmeɪʃn/
意思：自動化
解說：使用技術自動執行重複性任務的過程。
影片原句
And finally, capability number seven inside Codex is automation.
最後，Codex 內的第七項功能是自動化。
And finally, capability number seven inside Codex is automation.
最後，Codex 內的第七項功能是自動化。
延伸例句
Email automation saves time by sending messages automatically.
電子郵件自動化透過自動發送訊息來節省時間。
Email automation saves time by sending messages automatically.
電子郵件自動化透過自動發送訊息來節省時間。
invasive /ɪnˈveɪsɪv/
意思：侵入性的
解說：指介入或干擾他人隱私或空間的程度，此處指螢幕錄製功能可能讓人感覺隱私被侵犯。
影片原句
and this one may seem a little bit invasive.
這個功能看起來可能有點侵入性。
and this one may seem a little bit invasive.
這個功能看起來可能有點侵入性。
延伸例句
Some people find facial recognition technology invasive.
有些人覺得面部識別技術具有侵入性。
Some people find facial recognition technology invasive.
有些人覺得面部識別技術具有侵入性。
context /ˈkɑːntekst/
意思：上下文；背景資訊
解說：圍繞某事物發生的環境或資訊，幫助理解其含義。
影片原句
so it has context with what you're working on.
這樣它就能掌握你正在處理的內容。
so it has context with what you're working on.
這樣它就能掌握你正在處理的內容。
延伸例句
Please provide more context for your question.
請為你的問題提供更多背景資訊。
Please provide more context for your question.
請為你的問題提供更多背景資訊。
invoke /ɪnˈvoʊk/
意思：呼叫；調用
解說：啟動或執行某個程式、功能或命令。
影片原句
and the way that you invoke it is just say use Chronicle.
而呼叫它的方式只是說「使用 Chronicle」。
and the way that you invoke it is just say use Chronicle.
而呼叫它的方式只是說「使用 Chronicle」。
延伸例句
You can invoke the function by typing its name.
你可以透過輸入其名稱來呼叫該函數。
You can invoke the function by typing its name.
你可以透過輸入其名稱來呼叫該函數。
exponentially /ˌekspəˈnenʃəli/
意思：指數級地
解說：形容增長或變化速度極快，呈指數級數增加。
影片原句
and it's getting better exponentially.
而且它正在呈指數級改善。
and it's getting better exponentially.
而且它正在呈指數級改善。
延伸例句
The demand for cloud services has grown exponentially.
對雲端服務的需求已呈指數級增長。
The demand for cloud services has grown exponentially.
對雲端服務的需求已呈指數級增長。
句型解說（含實例）
if you use [A], but don't know [B]
意思：如果你使用 [A]，但不知道 [B]
接續：if + subject + verb + object, but + subject + don't know + question word + clause
解說：用於設定情境，指出使用者雖然擁有了某個工具，但缺乏使用該工具的知識或方法。
影片原句
You're in the right place if you use ChatGPT, Claude, Claude Code, Claude Co-work, or Codex, but don't know which one you should use for business and how.
如果你使用 ChatGPT、Claude、Claude Code、Claude Co-work 或 Codex，但不知道該在商業上如何使用它們，
You're in the right place if you use ChatGPT, Claude, Claude Code, Claude Co-work, or Codex, but don't know which one you should use for business and how.
如果你使用 ChatGPT、Claude、Claude Code、Claude Co-work 或 Codex，但不知道該在商業上如何使用它們，
實例
- If you use social media but don't know how to grow your audience, this guide is for you.
如果你使用社群媒體但不知道如何擴大受眾，這份指南適合你。 - If you use Excel but don't know how to use macros, you should watch this tutorial.
如果你使用 Excel 但不知道如何使用巨集，你應該觀看這個教學。
specifically, we're talking about [A]
意思：具體來說，我們將討論 [A]
接續：specifically, + subject + be + talking about + object
解說：用於將廣泛的話題縮小到特定的焦點或主題。
影片原句
Specifically, we're talking about Codex, OpenAI's agent tool.
具體來說，我們將討論 Codex，OpenAI 的代理工具。
Specifically, we're talking about Codex, OpenAI's agent tool.
具體來說，我們將討論 Codex，OpenAI 的代理工具。
實例
- Specifically, we're talking about the new marketing strategy.
具體來說，我們將討論新的行銷策略。 - Specifically, we're talking about the impact of AI on education.
具體來說，我們將討論 AI 對教育的影響。
break down [A] into [B]
意思：將 [A] 拆解為 [B]
接續：break down + object + into + parts/components
解說：用於將複雜的概念或系統分解為更易於理解的個別部分。
影片原句
In this video, we're going to be breaking down the seven core capabilities of Codex.
在本影片中，我們將拆解 Codex 的七項核心功能。
In this video, we're going to be breaking down the seven core capabilities of Codex.
在本影片中，我們將拆解 Codex 的七項核心功能。
實例
- Let's break down the project timeline into weekly tasks.
讓我們將專案時間表拆解為每週任務。 - We need to break down the budget into specific categories.
我們需要將預算拆解為特定類別。
by the end, you know exactly [A]
意思：到最後，你會清楚知道 [A]
接續：by the end, + subject + know + exactly + clause
解說：用於預示學習或觀看過程結束後，觀眾將獲得的明確知識或理解。
影片原句
so that by the end, you know exactly what it can do and where it fits in your workflow.
這樣到最後，你會清楚知道它能做什麼，以及它如何融入你的工作流程。
so that by the end, you know exactly what it can do and where it fits in your workflow.
這樣到最後，你會清楚知道它能做什麼，以及它如何融入你的工作流程。
實例
- By the end of the course, you will know exactly how to code in Python.
課程結束後，你會清楚知道如何用 Python 編程。 - By the end of the meeting, you will know exactly what the next steps are.
會議結束後，你會清楚知道下一步要做什麼。
not exactly [A]
意思：不完全等同於 [A]
接續：not exactly + noun/adjective
解說：用於澄清或修正先前的比喻或說法，指出兩者之間存在差異。
影片原句
At first glance, it kind of looks like ChatGPT or Claude, because you can create a chat, but they're not exactly the same.
乍看之下，它有點像 ChatGPT 或 Claude，因為你可以建立聊天，但他們並不完全相同。
At first glance, it kind of looks like ChatGPT or Claude, because you can create a chat, but they're not exactly the same.
乍看之下，它有點像 ChatGPT 或 Claude，因為你可以建立聊天，但他們並不完全相同。
實例
- It's not exactly a free service, but it is very affordable.
它不完全免費，但非常便宜。 - The result is not exactly what we expected, but it is still good.
結果不完全如我們預期，但仍然很好。
made to [A]
意思：旨在 [A]
接續：made to + verb
解說：描述某物或某人的設計目的或主要功能。
影片原句
Codex is made to do any task that you might want to do on a computer.
Codex 旨在執行你在電腦上可能想進行的任何任務。
Codex is made to do any task that you might want to do on a computer.
Codex 旨在執行你在電腦上可能想進行的任何任務。
實例
- This tool is made to simplify your daily routine.
這個工具旨在簡化你的日常生活。 - The software is made to handle large datasets efficiently.
該軟體旨在高效處理大型數據集。
realize that [A]
意思：意識到 [A]
接續：realize + that + clause
解說：表達突然理解或察覺到某個事實或情況。
影片原句
you'll realize that it is truly a super app that does any coding task or any knowledge work task.
你會發現它確實是一個超級應用程式，能夠執行任何程式碼任務或任何知識性工作任務。
you'll realize that it is truly a super app that does any coding task or any knowledge work task.
你會發現它確實是一個超級應用程式，能夠執行任何程式碼任務或任何知識性工作任務。
實例
- You will realize that practice makes perfect.
你會意識到熟能生巧。 - I realized that I had left my keys at home.
我意識到我把手機鑰匙忘在家裡了。
hop into [A]
意思：進入 [A]
接續：hop into + noun (often a topic, section, or application)
解說：非正式用語，表示開始進行某項活動或進入某個主題/部分。
影片原句
Let's not waste any time. Let's hop into capability number one.
我們別浪費時間了。讓我們直接進入第一個功能。
Let's not waste any time. Let's hop into capability number one.
我們別浪費時間了。讓我們直接進入第一個功能。
實例
- Let's hop into the next section of the presentation.
讓我們進入簡報的下一部分。 - I'm going to hop into the code editor to fix this bug.
我將進入程式碼編輯器來修復這個錯誤。