When everyone has AI and the company learns nothing
But we got CoPilot!
I stole that title from this great essay by Robert Glaser, which you should go read.
His key point is: most companies are doing this the old way (let’s make a deck, figure out a strategy, do some lunch & learns, roll out some access), and it doesn’t work. It’s too slow.
The dramatic coding breakthrough that happened last November hasn’t perculated through other disciplines yet, but the economics of it mean that it’s unstoppable. And the models will continue to improve for at least a few years and eat up much of the “work” that we all take for granted.
I’m finding it hard to convey the change to people without sounding like a weirdo. AI keeps getting better, and a lot of white collar jobs will change dramatically. I work with AI in the legal space, for example, you can see it happening there.
“It ain’t what you don’t know that gets you into trouble. It’s what you know for sure that just ain’t so.” — Mark Twain
Ethan Mollick has a great insights on this as well. If you’re a discipline leader in any organization right now, here’s what I think:
Create permission and look for surprises. Permission for individuals to try things. Some people in your teams will be well ahead of others. And someone who is ahead 3 months in AI time is 10 years ahead in real years. The frontier is moving very fast. You want to surface those people.
Build trust with other disciplines. Build real human connections and trust with engineering, with customer support, sales, you name it. Every discipline/specialization will have to change how they work, and at the same time other disciplines will change how they work as well.
The smart people from the past seem to have gotten this, so another quote to wrap up:
In the beginner’s mind there are many possibilities, but in the expert’s mind there are few.” — Shunryu Suzuki, Zen Mind, Beginner’s Mind
Interesting this week
The YC team can be a bit overly optimistic about AI, but this is interesting: how they built internal agents. Two interesting bits: 1. simply connecting an agent to your internal database (and having a central source of truth) is incredibly powerful. No need to make it more complicated than that. And 2., this stuff works well in high-trust teams.
Simon Willison notes that Anthropic's Claude Opus 4.8 release notes are refreshingly honest: "a modest but tangible improvement on its predecessor." (About their latest model.)
Patrick Neeman on UX Collective argues that designing for AI is like designing for the web in 1999 (no shit): no settled conventions, standards still being written as we use them, the interaction paradigm (chat) is just the first guess not the final form. I used to think chat was a temporary cludge, but my take has changed: I now think chat is much more useful than most UX people (us) tend to think.
If you haven’t followed what is happening in software engineering, Casey Newton on Platformer interviews Boris Cherny, the creator of Claude Code, about what happens to software engineers now that AI can code. (In short: hand coding is dead.)
Ethan Mollick's new research piece "Choosing to Stay Human" is about the risk of cognitive surrender: studies in Turkey and Taipei found students who used AI heavily showed measurable drops in independent reasoning.
Azeem Azhar explains why AI productivity gains aren't showing up in company results. He uses the electrification analogy: Stage 1 (lightbulb) speeds individuals, Stage 2 (group drive) speeds workflows, Stage 3 (unit drive) transforms the whole firm — but only if you rebuild decision-making, not just execution.
AI talks are getting booed at graduation ceremonies
Eternal Sloptember is just such a great title.
Remember - you got this!
Health and happiness,
Peter