Anthropic released a new feature yesterday.
It's called Claude Tag.
And I think when most people saw it, they thought,
okay, that's cool, a convenient way to invoke Claude
from within my Slack instance.
I use Slack, that sounds really cool.
But the more I read about it and the more I thought about it,
I actually started to get scared.
This is Anthropic's entry point into owning all knowledge work.
And this affects everybody, not just if you're writing code,
but literally if you do anything in front of a computer,
Anthropic is trying to own that.
And so I'm gonna show you what Claude Tag is
and then I'm gonna explain the bigger picture
of why this is such a big deal.
And by the way, if you wanna keep up
with what Anthropic and OpenAI are releasing
seemingly every week, like this video,
subscribe to the channel.
It really does help.
Thank you in advance.
All right, so let me show you what Claude Tag is.
If you use Slack, it's effectively
like just tagging another member of your team
and talking to it.
Except Claude Tag has all the context
of everything you do inside your company.
And it can actually get real world work done for you.
You simply tag Claude.
It chats with you.
It knows who you are.
It knows who your colleagues are.
It's able to understand all the different documents
you have inside your company.
It reads all the conversations in the Slack channels.
It is really building an entire graph of your company.
And by the way, you don't always have to tag Claude.
It's actively reading your conversations
in quote unquote ambient mode.
So again, it's just always sucking up
all the data from your company.
And Anthropic isn't mincing words
about what this product is.
They are literally saying Claude Tag is an evolution
of their cash cow Claude code.
Claude Tag is an evolution made more proactive
and built to work with a full team.
It's now one of the main ways we get things done
at Anthropic.
65% of our product team's code now comes
from our internal version.
So they're not even going into Claude code
directly anymore.
They are simply just typing from wherever they are.
And that happens to be Slack.
This is core infrastructure for Anthropic now.
And they want it to be core infrastructure
for every single company in the entire world.
And again, I really want to highlight
how big of a deal not only I think it is,
but Anthropic thinks it is.
This is Andre Karpathy,
one of the most prominent minds
in artificial intelligence who just joined Anthropic.
This is a new paradigm for interacting with Claude
that is significantly more in line
with all the other human activity org wide.
Interfaces are going away.
That is effectively what he's saying.
Don't go to Claude code anymore.
Don't go to any of the apps.
Wherever you're chatting,
which is Slack for a lot of teams,
that's where Claude tag is going to be.
And let me just pause for a second.
If Slack is where all the conversations are happening
and Claude lives in Slack,
what do you think Anthropic is going to build next?
I guarantee they are going to build a Slack competitor.
Okay, so let's continue.
Once you do all of the under the hood engineering work
to make this just work,
acts across tools, integrations, compute environments,
memory, security, et cetera,
Claude basically joins the team in a seamless way.
You can talk to it as you would talk to a person
and it can help with a very large variety of workloads.
They're positioning Claude tag
as another member of your team.
This is not a bot in your Slack app.
This is not AI responding to people tagging it
in your Slack app.
They really do say,
this is a new form of employee for your team.
And guess what?
You're renting it from Anthropic.
And he goes on to make even bigger claims.
In my opinion,
this is the third major redesign of the LLM UI UX.
The first paradigm was that the LLM is a website you go to.
So think about Claude or ChatGPT.
The second was that it is an app you download
to your computer.
Great, you can download ChatGPT,
but really what he's talking about
are apps like Codex and Claude Code.
And then third, a self-contained, persistent,
asynchronous entity with org-wide tools and context
working alongside teams of humans.
It really takes a while to wrap your head around it,
but it works and it is awesome.
And look, a lot of people gave Andre Karpathy criticism
for this because yeah, he joined Anthropic
and of course he's drinking the Kool-Aid.
He believes in what they're building,
but he also happens to be right.
This is a big deal.
You are now effectively hiring people,
people, AI, from Anthropic.
You have to pay them for that.
And at the same time that you're doing that,
you're handing them all the information about your company
and you are renting that back from them.
They are building an information graph
about every single company that uses Claude Tag.
And you need to decide if you're comfortable with that.
More on that later.
And by the way, it seems like I'm going to be paying Anthropic
more and more.
And to help me fund that,
let me tell you about the sponsor of today's video.
Context engineering is an incredibly important part
of your AI stack.
Making sure your AI has all of the data it needs
to give you the best possible results
is actually kind of a hard problem.
And if you're just working in one document,
sure, just give it that one document.
It's easy.
But imagine if you have thousands of documents
and then you start mixing in videos
and articles and other types of media.
And as you're seeing from Gbrain
and from Karpathy's LLM Wiki,
people are building all of this stuff themselves.
But it is not easy.
That is where Recall 2.0 comes in.
Recall 2.0 allows you to throw everything
into this kind of ocean of data.
And it can provide the right context
at the right moment to your AI.
With Recall 2.0,
you can switch between chatting with your knowledge base
and researching on the web.
I can add all of my OpenClaw videos
to my Recall knowledge base,
ask what the best security practices are,
and Recall will even give me timestamps
that I can watch in the app.
With an API and MCP,
you can plug this into any system
that you're already using and building.
And the best part,
you're not locked into one model.
So go check out Recall.
It's a great product.
Use MB25 to get 25% off.
And I'll drop the link down below in the description.
Thanks again to Recall.
Now back to the video.
And by the way,
Y Combinator has been talking about
this kind of AI native company for months now.
And they've put out videos,
but nothing really concrete about how to build it,
what it looks like,
companies that are already using it.
But they've been talking about it.
And now this is what they're talking about.
This is the beginning of the AI native company,
where AI understands every inch and corner
of your company
and is able to run 24 hours a day,
helping you grow,
finding issues,
helping your human employees
be a thousand X more productive.
This is the vision
that they've been laying out for months now.
And of course,
Anthropic has been building it.
This is Ankit Gupta,
a general partner at Y Combinator.
I like that at YC,
the partners and software team
have consistently built internal versions
of pretty much every product I see
open AI and Anthropic launch
six to 12 months later.
Let's see around corners
for where things are going.
So that's what I'm talking about.
They've been talking about AI native companies,
but they haven't put out
any public facing products
and they certainly haven't open sourced anything.
And I'm going to point back
to this incredible essay from Anthropic
because this is effectively
what they're trying to do
for all knowledge work in all companies.
Now, this is the essay
in which they talk about
self-improving artificial intelligence.
I'm not going to talk about it again.
I already made a video about it,
but what we're effectively seeing
is what happens
when a company becomes AI native,
when AI is fed
every single piece of data
about a company,
about all the conversations happening,
all the processes,
all the documents,
all the emails, everything.
It's all of a sudden
able to have this incredible visibility
into what the company is doing.
And then you can set goals.
I've been talking a lot
about loops lately.
Imagine if you can set a loop
for all knowledge work.
The loop can simply be
increase the revenue of my company
and it is going to ingest
all of the context,
all of that understanding
of the company,
set up experiments,
and then loop over it indefinitely.
And if that sounds expensive,
it is.
It is going to take
basically infinite tokens.
And then we're at this point,
and this is where it really gets scary,
where whichever company
can afford the most tokens,
they're going to do the best
because AI is going to improve
their business the most.
This sure sounds like the beginning
of the permanent underclass.
And then there was this tweet
by Ashwin Gopinath,
and he said it perfectly.
Claude Tag is a Trojan horse,
and I could not agree more.
Not because Anthropic
is doing anything evil,
because the incentives are obvious.
So day one,
this looks like a great feature.
Tag Claude in Slack,
let it follow the thread,
remember context,
connect the tools,
break down tasks,
chase work,
and act like a teammate.
But that's exactly the problem.
The moment your AI vendor
becomes a shared coworker,
it stops being just a model provider.
It starts becoming the place
where work is interpreted,
remembered, routed,
and eventually executed.
Imagine if you no longer
had any employees on your team,
and you were renting
an AI employee from Anthropic.
Now imagine that
across the entire economy.
Every single piece of knowledge work
is now just rented
from this one massive company.
That's why I keep saying
this is scary.
So that is not model lock-in,
that is context lock-in.
You are now renting your company
back from them.
And yes, platform risk is very real.
It was real before you could
literally buy intelligence
from a company.
It was real when people
were building apps
on top of the Apple App Store
and Apple got to decide
if they liked your app enough
to let you be there.
And whether you were building
on top of Facebook
and they wanted you
to use their API or not.
Anytime you build
on top of somebody else's platform,
you have platform risk.
And this is the ultimate form
of platform risk.
It is all knowledge work
platform risk.
He goes on,
the pricing model
makes it even more dangerous.
A human coworker has a salary.
Claude has unbounded
tokenized activity.
That means once again,
there is a cap
that you could pay somebody,
a human worker.
But if you're paying Claude,
it is uncapped.
You can literally pay them
infinite dollars
and they'll eat it up.
And there's always
more work to be done.
Imagine running a company
and being that dependent
on another company.
But it actually gets bigger
and worse than that.
What happens when
every single company
starts building
on top of Anthropic?
And I know I'm talking
a lot about Anthropic,
but that's because
they released Claude Tag.
But you know OpenAI
is working on the same feature.
Hopefully other companies as well.
I hope there's more competition
because competition
will be the only solution
for the problems
that come from Anthropic
owning all knowledge work.
So imagine now
you're a SaaS company.
You have software.
You sell it to other companies.
But all of a sudden,
Anthropic is now
using their agents
to operate your software
on behalf of their customers
that pay them.
And guess what?
Those customers
are no longer
going back
into your user interface.
They no longer log in.
They have no need to.
They just tell their agent
what to do.
So why would they ever?
All of a sudden,
your user interface
becomes valueless.
And so what are you
at that point?
You have the workflows.
So remove the UI.
What can an agent
actually get done
inside of your application?
Those are the workflows.
But how long does that last?
Because really,
the agent can just write code
for those workflows.
AI is extremely good
at writing code.
So why wouldn't it
just write code
to manage
and then eventually
overtake those workflows itself?
As a software company,
what are you at that point?
You're simply a database.
And you know what's easier
than writing code
for workflows?
It's writing code
to read and write
from databases.
Agents are incredibly
good at that.
They don't even care
what kind of database it is.
They will help you
move your data
from one database
to another.
They'll help you read
and write from it easily.
And then all of a sudden,
what happens
to all software?
No software
is safe from this.
Agents are now
infiltrating
right before our eyes
every single piece
of software.
That brings us
to our final
scariest point.
Anthropic all of a sudden
owns all knowledge work.
They own
all software.
How does society
continue at that point?
And that's the problem.
Government intervention
is the last line
of defense.
Societally,
we will not allow
a single company,
a single entity
to have all the power.
It just won't work.
But there is
a solution
beyond that.
Open source models,
model competition
from other research labs,
making sure that
whatever provider
you choose
for this Claude
tag-like feature
allows you to
own the context.
You own the context
of your business,
not them.
You should own it.
And that's why
open source
is so important right now.
You're not going to be
beholden to paying
Anthropic prices
for every token.
You can pay
the best price
for the best token
when you have
a multi-model,
multi-provider strategy.
So what started
as a cool,
convenient feature,
being able to tag Claude
from within Slack
is very obviously
Anthropic's attempt
to own
all knowledge work.
That is a very bad thing.
But the one thing
I do want you to remember
and think about
is this is
definitely the future
of how companies
will work.
AI and humans
working in concert
to accomplish things
and AI
living and breathing
inside of your company,
not just
a chat app
that you go
type a query
to get the response
and then continue
your work.
It is going to be able
to see everything
you're doing
inside that company
and it's going to be able
to help you.
It's going to be able
to make you
so much more productive.
So I wanted to end
on that optimistic note.
We just need more competition
most of all.
And I talked about Anthropic
having self-improving
artificial intelligence.
I actually made
an entire video
breaking that down
in detail.
Go check it out
right here.