AI Gaps: Opportunities for Humans
Key Points
- The “job families at risk” framework is outdated for the AI era; instead, we should focus on identifying human talents that fill the obvious gaps where AI still falls short.
- AI excels in many tasks but remains weak at fuzzy‑logic activities such as competitive assessment, sales intuition, and go‑to‑market strategy, leaving those strategic roles in high demand.
- Complex interaction design is another clear blind spot for AI—its outputs tend toward overly simple, single‑button interfaces, while nuanced, multi‑step user experiences require deep human expertise.
- Enterprise‑level solution architecture cannot be reduced to a “dumb” role; senior architects are still essential because AI lacks the holistic reasoning and training data to design robust, end‑to‑end systems.
Full Transcript
# AI Gaps: Opportunities for Humans **Source:** [https://www.youtube.com/watch?v=1Jve3NIIAuY](https://www.youtube.com/watch?v=1Jve3NIIAuY) **Duration:** 00:09:18 ## Summary - The “job families at risk” framework is outdated for the AI era; instead, we should focus on identifying human talents that fill the obvious gaps where AI still falls short. - AI excels in many tasks but remains weak at fuzzy‑logic activities such as competitive assessment, sales intuition, and go‑to‑market strategy, leaving those strategic roles in high demand. - Complex interaction design is another clear blind spot for AI—its outputs tend toward overly simple, single‑button interfaces, while nuanced, multi‑step user experiences require deep human expertise. - Enterprise‑level solution architecture cannot be reduced to a “dumb” role; senior architects are still essential because AI lacks the holistic reasoning and training data to design robust, end‑to‑end systems. ## Sections - [00:00:00](https://www.youtube.com/watch?v=1Jve3NIIAuY&t=0s) **AI’s Gaps: Human Opportunities** - The speaker argues that instead of labeling entire job families as “at risk,” we should focus on AI’s clear weaknesses—such as strategic intuition, competitive assessment, and complex interaction design—where human expertise still provides essential value. ## Full Transcript
you need to
realize that job families at risk is the
wrong category for assessing human
talent in the AI era we talk about it
all the time Chad GPT 40 came out with
that image update thing this week my my
DMs are full of graphic designers and ux
designers worried about their jobs even
agencies are worried
fine what we're
missing is that AI as it advances is is
leaving really obvious cracks like if
you look at AI as this advancing uh wall
of intelligence which sounds really
scary right what you're seeing is that
it's not really a wall it's actually
like full of canyons and cracks and
areas where it's not getting better and
the more certain parts of AI improve the
more obvious it gets where those gaps
are let me lay out a few that I think
are incredibly obvious as human
opportunities and they're not getting
better one is figuring out how to do
competitive assessments how to do fuzzy
logic tasks around competition sales go
to market distribution as a
whole that is not being close to solved
with AI I I don't just mean that AI
can't sit down for dinner and close a
deal that's certainly true and people
have said that before but I also mean
that AI is really bad at the gut level
intuition that comes with good
distribution and good distribution
strategies and not a lot is getting put
into that another
example AI is really really bad at
complex interaction
design if you ask it to think in terms
of a complex interactive system things
that you do with a complex app it's not
good at it like when I talk about 40 and
designers being worried in my DMs I look
at what it actually produces it's a very
simplistic app it's like an app with a
single
button and if your app has a single
button you're probably not going to last
anyway you're not going to make it you
need complex interaction design and
that's something that by definition AI
can struggle with because it tends to
think in terms of next token prediction
and systems design thinking is
inferred through things like reasoning
models you can get to completeness of
text through things like 01 Pro but
interaction design is not text it's also
not code it is a third thing and it is
not something that the models have great
training data on and they're not
necessarily getting better at
it here's another
one technical solution architecture is
something that enterprises used to have
and it was worth doing because
Enterprise
could invest in an entire software
engineering team to carry out the
architecture once it was put through by
someone who was very senior and
understood the systems blah blah
blah okay
fine now everyone can make software
everyone has a team of like kind of dumb
intern engineers in their
pocket we have no dumb solution
architect
equivalent uh it's not there because you
can't have a dumb solution Architect by
definition and another category of DMS
that I get all the time is what do I do
to design this app now that I've built a
splashy looking front page in lovable or
in bolt or whatever it might be I have a
nonfunctioning homepage isn't that
impressive well they know it's not and
they want it to be better but they
didn't go to school for database
architecture and they're not going to
and so there's this enormous opportunity
opening up for figuring out how you
bridge that Gap now if you're going to
scale that out to the millions of people
who now want to build small apps there
will be some productization there I
don't I I I have no
Illusions
but there is also increasingly a human
component to this because you're not
just thinking in terms of what is the
technical system for the idea you
brought to to me you have to use human
fuzzy logic and figure out what is your
inferred intent long term and then what
is the right tool selection and then
within that tool selection what is the
right sequence of steps overall to
prompt the tool with to help it help it
build these are things that we don't
really have names for but they're highly
important
roles here's another one chat GPT
now needs to be something that marketers
measure and understand when they look at
brand profile and they look at mentions
and all of that nobody knows how to do
that nobody knows how to game that and
like get more chat GPT
mentions that's something that yeah
maybe there's some science there maybe
stripe launching lm. text is going to
help their visibility in future chat GPT
updates but at the end of the day it
becomes a human problem to figure out
how to optimize the web across multiple
interfaces the web is being
disintermediated it used to be human and
computer and now it's human computer
it's human chatbot and it's human agent
which may choose a computer or choose a
chatbot and for marketers trying to
solve that problem like that's a whole
new class of problem to solve for
reaching people ads traditionally don't
don't work like does your ad work on an
agent have you tested it that's a human
problem and so I'm not just saying we
are thinking about the wrong thing when
we think about Job families disappearing
because I'm an optimist and I'm not
saying it because we shouldn't actually
spend some time thinking about it all
I'm saying is the pendulum has swung so
far over on the Gloom and despair side
that we are ignoring these really
interesting niches that are coming and
that are becoming more obvious as
intelligence is scaling like one of the
things I look at is where is the
direction of these intelligences going
where is AI evolving
faster AI is evolving on medical and
science Innovation really fast AI is
evolving on code production really fast
AI is evolving on complex text
production for non-fiction purposes
really fast AI is evolving on reasoning
across the we web really
fast but there's a lot of other
categories where it's not really moving
much at
all and so we need to think about what
that looks like we need to think about
areas where we are not seeing those
Investments and what it means from a job
family perspective to refocus and
rethink what jobs are meaningful in the
AI era because Claude can be great at
code production but if you don't have
that human basically able to to say this
is where the code should should be
written this is the language it should
be in this is how it should talk to the
database this is why this is how that
maps to the inferred intent of the app
designer you're not going to get very
far with claw and and that is the
Baseline experience of so many people
who set out hopefully to build apps with
these tools and I know there are smart
people at lovable. deev and at bolt. new
trying to figure this out repet trying
to figure this out
great but the challenge is you have to
field an entire range of human building
utterances with this and infer from that
in systems thinking reliable
architectures that are secure and
oneshot apps that is a very hard
Challenge and I'm not saying it's
impossible but I'm saying that it leaves
a lot of room for smart humans who can
help with technical allocation
challenges
so we have technical allocation
challenges we have marketing challenges
arguably product challenges we have
engineering system design
challenges we have uh sales distribution
challenges there are things in most of
these job families that remain very
unsolved and that AI is not getting
better
at so think about that
and have some hope even though like a
new thing comes along and Sam Alman does
a l we Hill your startup
slide think about where job families are
going and look at how you can upskill to
in that direction cheers