Week 3: Big deal, and only a big deal

The biggest AI news this week was paperwork. OpenAI filed to go public on May 22, Anthropic is lining up for the autumn, it's a huge bet for the whole American economy, but what I'm mostly looking forward to seeing is not the financials but the usage numbers. The narrative has been about huge adoption and independent estimates seem to confirm, but hopefully finally we'll have a sense of the scale.

One interesting take about this was the one by Benedict Evans on Lenny's Podcast. His take is that "AI is as big a deal as the internet or mobile, and only as big a deal as the internet or mobile". Huge, but not magic. It helps to keep both halves in mind and the week's news stops being flashy apocalyptic headlines and turns into a set of receipts for the same boring, useful idea. It made me think about where all of this is going and I wrote it up in full on the blog. Here are three of them:

Where the model lead actually lives

OpenAI had the first mover advantage with ChatGPT, then for months Claude Code was the one to beat, top of the benchmarks that matter and writing a real slice of the world's code. Then OpenAI put out a Codex release at the end of April, and within a week it had passed Claude Code on downloads. Next release, it could swing back.

When the lead changes hands that fast, "moat" is the wrong word for what the models have. They don't lock you in the way Windows did; there's always a comparable one a click away. Evans puts the sharp version of it well: run the same prompt through the big models blind, and most people couldn't tell you which one wrote which. The durable advantage is going to live one layer up, in distribution, in the product, in being the tool people already have open. Which is why the labs are racing to be the default rather than the smartest.

What's actually happening to the jobs

The loudest take of the year is that the labs are about to fire everyone. The receipts say something quieter is happening first. The labs themselves are hiring as fast as they can fill seats, OpenAI is on track to roughly double its headcount this year, and the consultancies are having a boom selling AI to everyone else. It is also true that some of those same firms are cutting in other places at the same time. Work is moving from one column to another, the way it did when the spreadsheet arrived and the number of accountants somehow went up through fifty years of software built to replace them.

Companies are starting to meter AI

Uber just capped what its engineers can spend on AI coding tools at $1,500 a month each, after some of them turned internal token-spend into a leaderboard sport. Simon Willison did the math: that cap is around 11% of a median Uber engineer's pay, which makes it a sustainability line rather than a punishment.

The signal underneath it is that the individual subsidies are running out. For two years the per-seat prices hid what this stuff costs to run at scale, and the 2025 budgets were written before anyone knew how hungry the 2026 agent tools would get. Companies metering their AI spend is what real adoption looks like once it's big enough to show up on the bill.

The posture under all of it

The thread through these three angles is one: presume radical uncertainty. Take it completely seriously, use it every day, and still refuse to pretend you know which of these bets pays off. Evans's own motto is better than anything I'd write for it: probably going to be okay, not for sure. The labs going public will hand us real numbers soon, and my guess is they'll come back messier and more boring than either the believers or the doomers want. That tends to be where the truth is.

---

Thinking about building something? If you want a technical partner to figure it out with, reply to this email and tell me about it, that's the kind of conversation I like having.

---

You're getting this because you signed up at christianvismara.com/newsletter. Forward it to someone who'd argue with it.

Past issues are in the archive.

Christian