Intelligence Saturation and Job Replacement
Sections
- Intelligence Saturation Limits AI Tasks - While AI can now perform individual tasks so well that further model upgrades offer diminishing returns, jobs demand sustained intent over long periods—a capability AI still lacks.
- Reducing AI Workflow Friction - The speaker highlights how seamless AI integration—beyond simple chat—can cut copy‑paste overhead, noting that code tasks benefit from clear metrics while document tasks lag, urging builders to focus on deeper workflow integrations.
Full Transcript
# Intelligence Saturation and Job Replacement **Source:** [https://www.youtube.com/watch?v=d4dmkm_58SM](https://www.youtube.com/watch?v=d4dmkm_58SM) **Duration:** 00:06:22 ## Sections - [00:00:00](https://www.youtube.com/watch?v=d4dmkm_58SM&t=0s) **Intelligence Saturation Limits AI Tasks** - While AI can now perform individual tasks so well that further model upgrades offer diminishing returns, jobs demand sustained intent over long periods—a capability AI still lacks. - [00:03:16](https://www.youtube.com/watch?v=d4dmkm_58SM&t=196s) **Reducing AI Workflow Friction** - The speaker highlights how seamless AI integration—beyond simple chat—can cut copy‑paste overhead, noting that code tasks benefit from clear metrics while document tasks lag, urging builders to focus on deeper workflow integrations. ## Full Transcript
Have you heard of the term intelligence
saturation? That's the idea that AI is
getting so good that it saturates the
task. It does the task so well that
further intelligence gains. There's no
point. What's what's the point? That is
already happening with some tasks. And
it's happening at the task level, not
the job level. And I want to lay that
out for you. So I talked to someone this
week who said, I know 03 is supposed to
be better as a model from chat GPT. I
don't see it. And I don't see it because
the task that I do 03 is not better at.
I'm already oversaturated on
intelligence. And that's different from
saying that the 40 model or whatever
they're using, I think they're also
using Gemini, is good enough to take
their job. Because if you think about
it, a job is so much more than a task. A
job is the ability to maintain intent
over time, for example. That's not
something AI is very good at. And to be
honest with you, I don't see a path
there super soon. Even our best
estimates, our optimistic guesses around
how long agents will maintain intent for
look like roughly a week, maybe a couple
of weeks, in the next couple of
years. We are not going to see a point
in the next couple years where AI is
going to maintain intent over years.
That is something that humans do really
well and that is an important part of
our job.
I have not seen any major model maker
claiming that an AI can maintain intent
for two years in the next two years. I
don't see it. But that's about the
average tenure of a tech job. Now, you
can argue a lot of knowledge workers
also don't maintain intent over two
years. And I suppose that's a fair
critique. But the point is we can still
be saturated at the task level before we
get close to job level replacement. And
that is what I see starting to happen. I
currently feel a difference with 03 that
is remarkable because of the kinds of
tasks I'm doing with it. It just works
better for what I ask it to do. I can't
tell you because it's hard to see coming
up which model coming down the pipeline
from a major model maker like Google or
OpenAI is going to be so good. I can't
tell the difference anymore. But I
believe it's coming. I believe we are
getting to a point with intelligence
where it will saturate out and we will
have all the intelligence we need for
our tasks and I think it's coming very
very soon much sooner than general
intelligence and if that's the case we
need to talk about where advantage lies
what do you do when everyone has the
intelligence the intelligence has
saturated the market and really the only
thing left is your ability to maintain
intent over time to do the things that
are unspoken in the job like Pan's
paradox. What's what's left from an AI
perspective that gives you a comparative
advantage? If you're a model maker, it's
a commodity. If you're an app builder,
it's also a commodity. Where's the
advantage? The advantage, I think, lies
in the integration into the tool chain
and the workflow. The advantage rolls
into how you install the intelligence.
So it makes it easier to get work
done. And that may be a company level
thing where large companies think about
how they integrate AI into workflows. It
can be an app level thing where apps
think about how they are designed for
specific workflows with intelligence in
mind. Because if you have to choose
between equivalent intelligence, aren't
you going to pick the one that makes it
easier to actually get the job done? And
I will say there's a lot of room to grow
here. We may have task level
saturation, but me and everybody else I
know who uses AI a lot complains about
the degree to which we are copying and
pasting stuff back and forth and arguing
with our AI and arguing with a different
AI and then copying it back over
somewhere else. There's a lot of
overhead associated with managing an AI
that we have not managed to get rid of
or come close to addressing. And that's
just the easy case where you're just
chatting with it. That doesn't come
close to talking
about integrations that dive more deeply
into workflows, document approvals,
document reviews, code reviews, things
like that. We are getting farther on
code, by the way, than the other areas
because code has such a clean reward
system. The code runs or it doesn't. The
code is clean or it isn't. Very, very
easy to measure. other areas like
documents in text are trickier to
measure and so that we're seeing slower
progress there but we will see progress
and I actually think that's one of the
key areas for builders to get an
advantage as intelligence starts to
saturate. So the purpose of this video
has basically been to introduce the idea
that we are getting to intelligence
saturation faster than we are getting to
general intelligence and that that is
going to matter because it shapes the
competitive landscape profoundly. It
means that we are going to see a lot
more people like my friend I talked to
last week who just say, you know what,
it's good enough. It's good enough. I
don't care when the new model comes out.
It doesn't matter. It's good enough.
We're kind of there. If you if you
wonder what that feels like, we're kind
of there with with the iPhone, with
mobile phones. Yeah, a new one comes
out, a bunch of people buy it because
they want the new thing, but it's not
revolutionary. That device is kind of
saturated. And that's why Tim Cook is
reportedly spending so much time
thinking about the new device. He's
apparently entirely obsessed not with
AI, but with a glasses-like device that
will enable you to look at the world and
compute. We'll see how that
goes. The point here is that saturation
is here and is happening and the tide is
rising on saturation. And if you don't
feel saturated yet, I don't feel
saturated. You are going to soon. You
may feel saturated by the end of this
year. In fact, I might be like 04. I may
be like, it's hard to tell the
difference with 03. I don't know. By
'05, maybe it's hard to tell the
difference versus
04. Keep your eye out for that. That
doesn't mean that you're dumb. It just
means that for the task that you're
using it for, there's not a meaningful
difference.