Your newsletter digest — May 29, 2026

Two pieces landed in your inbox since yesterday's digest: Ben Thompson wrapped up the week with a conversation about AI models and what advertising reveals about human nature, while Lenny Rachitsky put out an unusually dense reading list — 36 books organized by the skill you're trying to build.

Source: Stratechery · Published May 28, 2026

Ben Thompson's Thursday interview with Eric Seufert 1 covers three threads that don't often get braided together:

Why building AI models is harder than it looks. Seufert — a mobile marketing analyst who has closely tracked Meta's ad infrastructure — has been watching what it takes to actually construct generative AI models at scale. The interview draws on his view that the engineering difficulty is underappreciated outside a small community of practitioners.

Why Meta's foundational models matter more than their benchmarks. The conversation positions Meta's open-weight releases (the Llama family) not just as a strategic move against closed frontier labs, but as the structural foundation that allows a very different kind of AI ecosystem to exist. Seufert's advertising background gives him a lens the typical AI analyst lacks: he sees Meta's open model strategy as consistent with how Meta has historically subsidized the broader app economy to keep its own ad business healthy.

Why understanding advertising is grounds for optimism. This is the most counterintuitive thread in the interview. Seufert argues that advertising — typically framed as manipulative or extractive — is actually evidence that humans can be persuaded to want things, which means human preferences are malleable and not fixed. That malleability, in his reading, is what makes AI's upside so large: if what people want can shift, the space of possible futures is genuinely open.

The full interview is behind Stratechery Plus. 1

Source: Lenny's Newsletter · Published May 26, 2026

Lenny Rachitsky published a reading list 2 organized not by genre but by what you're actually trying to fix — a small but important difference from most recommended reading lists. His constraint: three books per category, and only books he's finished himself.

The public section covers nine skill areas:

Lenny's framing note is worth keeping: he follows Marc Andreessen's rule of reading books older than 10 years, because those are the ones that survived long enough to be tested. Most of the list holds to that — the exceptions (like Continuous Discovery Habits and Scaling People) are recent enough that their staying power is still being earned.

A category on productivity is behind the paywall, along with a promised part 2. 2

Both pieces today are, in different registers, asking the same question: what do humans actually want, and how durable are those preferences?

Seufert's optimism about AI rests on the idea that wants are malleable — advertising proves this. Lenny's reading list rests on something adjacent: that the craft of building, communicating, and leading has been stable enough that books written decades ago still hold. The tension between those two views — human preferences as fluid vs. human nature as settled — is probably the most interesting lens for thinking about what AI will change and what it won't.