AI Links, 7/18/2026

Arvind Narayanan on gradual diffusion; Andrey Mir on the digital world vs. the real world; Kai Williams on the Tokyo coding tournament; Hollis Robbins on shape rotators

Interviewed by Yascha Mounk, Arvind Narayanan says,

any task in the economy that you can specify precisely enough for it to even be legible as a task is going to be done by AI. …What jobs will mean in the future is what economists sometimes call interstitial tasks: the things that lack a precise specification, where maybe part of the task is even to figure out what should be done next, things that maybe have never been done before. Those are the kinds of things we think will be in the domain of humans. The reason I think we will be very unlikely to leave those things to AI is because of unknown unknowns. You can’t be sure that when you task AI with doing something so different from anything that’s been done before, it’s going to do a halfway reasonable job.

He points out that even now, a lot of our jobs do not serve basic needs.

In affluent societies, 95% of jobs are things we do because we need things for people to do, not things we do for people to survive. I’m saying that in the future, that 95% will go to 99%, maybe even 100%.

…what actually has value, it’s just very contingent on what happens to be scarce at any particular point in history. Things like light and clothing—things at the bottom of Maslow’s hierarchy—used to be extremely scarce because they were hard to produce, and they employed a very large fraction of the workforce.

…I think a lot of the things people do today are scarce and valuable precisely because human labor is scarce. I disagree with the premise that in the future AI systems will be creating most of the value. That’s not consistent with my understanding of economics. I agree that AI systems will be doing most of the work. But any work AI can do is infinitely reproducible, and when it’s infinitely reproducible, it doesn’t have a lot of value, because its cost is going to come down to the market—anybody can provide it at very low cost. So the value is going to shift to the kinds of things that are scarce.

One reason he thinks that the impact of AI on business will take longer than expected to materialize:

CEOs often build prototypes of some product and think, this is amazing, Claude Code built for me in twenty minutes what it took a team of two people two weeks to produce. But what the CEO doesn’t see is the last 90%, or even the last 99%, of what it takes to go from that prototype to an actually functioning, maintainable product. That’s what the line-level employees deal with every day. A lot of this information is not floating up from everyday workers, who actually have a much better perspective on these things

Andrey Mir writes,

Projecting medieval scarcity onto a technological future doesn’t hold up because the level of technological sophistication required to make such missions viable would simultaneously make their stated goals obsolete. By the time mankind could routinely mine asteroids or colonize distant worlds, it would also be able to synthesize virtually any material and, more consequentially, transcend biological existence altogether into digital form.

You don’t cross the cosmic ocean for iron ore or black pepper when you can print it.

Kai Williams writes,

unlike humans, who can only solve problems one at a time, OpenAI ran separate agents to solve every problem in parallel. Minaiev acknowledged that this gave OpenAI an advantage. But he argued that the AI solved the first four problems quickly enough that it would have won even without that advantage.

He describes a coding tournament, won by OpenAI. But then he circles back to Narayanan and a co-author, who see software development as a “sandwich,” with coding in the middle while requirements development and deployment are on either end.

Today, AI systems are mostly useful in the middle of the sandwich — execution — which is exactly what competitive programming focuses on. Competitors are given well-specified tasks to solve, and ultimately, all they need to deliver is code (no matter how messy) which passes automated tests.

So even if AI systems are well beyond the best humans at competitive programming, it doesn’t necessarily mean that they’ll be able to do tasks humans currently do in other parts of the sandwich. For instance, AI agents can’t yet organize a focus group to learn about the needs of real users, nor can they negotiate with colleagues about the best way to deploy a new version of their software.

Hollis Robbins writes,

Aristotle had his day. His premises (and Ben Sira’s and Burke’s) were that wisdom requires leisure and that verbal skill is scarce. It used to be hard to produce a scribe. There were fewer Jeffersons than Shermans back in the day, which is why Jefferson got the main drafting job. For two millennia nothing dislodged the primacy of verbal ability, until now. Now shape rotators may very well rule the next era, though wordcels will still be needed to create prestige TV as this era’s circuses. I’ve advocated that students should be able to go to college to study English and then go be a plumber. Jonathan Rose’s great 2001 book The Intellectual Life of the British Working Classes is all about not assuming anything about the civic virtues of those who work with their hands.

And so I’m bearish on the knowledge economy and bullish on rotating shapes.

substacks referenced above: @

Let Us Now Praise Famous Shape Rotators

@

An OpenAI model crushed top human programmers at a world coding competition

@

Cascade of Singularities

@

Arvind Narayanan on Why AI Isn’t All That Revolutionary