It’s a paradoxical moment in the great global AI rollout. At one level, you have three of the largest IPOs in human history headed our way: SpaceX, which rolls up Grok, as well as OpenAI and Anthropic. It’s a deca-trillion-dollar ouroboros economy predicated on the cost of Nvidia chips. 

At the same time, certain narratives are fizzling out. “Tokenmaxxing”—encouraging engineers to burn as much AI time as possible—seems to be running its course. OpenAI’s economics are shaky pre-IPO, at least given its proposed valuation. AI-assisted coding, like Codex or Claude Code, seem to be driving the most growth, but no other application of AI is quite as profoundly disruptive. And we keep learning that most AI projects fail. 

Even the idea that AI is taking all the jobs is fizzling out. Dan Shipper at Every.to recently published a long exploration of how AI is increasing the amount of work people have to do (while also letting them do more)—we keep seeing the same, and while the number of junior employees being onboarded is shrinking, the overall rate of engineering hiring seems to be going up, even as AI gets more powerful. Sam Altman says he’s “delighted to be wrong” about AI taking all jobs, which—well, I’ll let you unpack the ironies while I breathe into a paper bag. Meanwhile, Erin Brockovich (yes that one) is making maps of data centers, and not because she loves them. The Pope has thoughts, too.

My guess is we’re at, or close to, peak AI hype. Silicon Valley is very close to its big AI payday, which unfortunately is going to lead to some of the most exhausting people on earth having the most money on earth, and will transfer all the risk of this great experiment into the public markets. Not optimal, but this is the society we have built. In a year or two, we can go back to talking about literally anything else on our podcasts. I think it’s safe to start planning your post-hype lifestyle. For me, in addition to building our AI-focused software consultancy with Rich and our amazing team (which is work I love), I plan to reread Middlemarch and focus on climate models again, inshallah. Maybe more bike rides.

I think at moments like this, it’s worth taking a breath and noting what you actually have observed with your own eyes. Here are three things I see and believe:

1. With an experienced software architect in the loop who has a good process around AI, you can build software far, far more quickly than before, especially big “boring” software oriented around lots of database transactions. We know that software is a multi-trillion dollar industry with enormous amounts of process, and, at least for me, its fundamental methods are different than they were a year ago. Lots of software that no one could afford to rebuild is now so affordable to build you could start over from scratch. But without that experienced architect who is good at AI stuff, you probably will just make a mess and your “pilot” will fail.
   
   
2. Notice I emphasized rebuilding. Building good new “product”—actually making software that makes sense to humans, is easy to use, and is not like most of the software that came before—is about as hard as before. In some cases, it’s harder, because there’s so much more stuff getting produced, and AI is a slippery substrate. Product managers being able to code themselves doesn’t radically change this as much as people might have expected. Imagine an English teacher asking people to turn in their five-paragraph essays and everyone turns in a bad novel. That’s what product managers are dealing with in the era of vibe coding.
   
   
3. You can’t really trust AI outputs yet. If you’re going to use AI for knowledge work, you have to be very paranoid; attempting to hide from that would be dumb. Ethan Mollick disagrees, but for me, I don’t feel safe relying on those outputs. It’s good at summarizing things, though.

I’m curious what you believe; feel free to get in touch and share.

Do those things add up to trillions in value? Probably. Are the companies that exist today going to capture the trillions? Probably not. Ultimately, I don’t care that much about that part. I’ve been through this cycle before, many times. As the IPOs get closer, the promises get more and more extreme. Greed settles like a haze over everything. 

Everyone expects the bubble will pop—they just hope it pops after their IPO. And then it’s like Times Square the morning of January 1: Everyone has gone home and there’s confetti everywhere and people pushing brooms. I have never been to Times Square for New Years Eve, because it seems terrible. But I love going to Manhattan the morning after—especially on the earlier side—and walking around. Everything is emptied out: You have the whole city to yourself. And you can look around and think about how you want the next year to go.