Hello,

It's been a slow week here. The weather finally turned — first proper warm weekend of the year — and Taylor has decided that the rug by the open patio door is the best place she's ever been. I'm inclined to agree.

I wrote something this week that's been sitting with me longer than usual. Emmett Shine said something a while back about how AI output always lands between a 2 and a 7. Never truly bad. Never truly great. Just somewhere in the middle. The line that's been stuck in my head: AI is bad at making things that are bad.

That part doesn't get discussed much. We've all heard that AI can't make a 9 or a 10 — taste, judgment, all of that. Fine. But the floor is the more interesting half. AI raises everything to a 5. Which sounds good until you remember that most professional work has lived around a 5 forever. And that almost everyone I know, myself included, got to whatever they're good at by making a lot of bad stuff first.

​Read the full piece →​
​
Until next time,
Anton

Worth your time

• ​After Automation — Dan Shipper at Every makes the case that AI doesn't reduce work for experts, it relocates it. Sits right next to the 2–7 piece. Long, but the long ones are usually the ones worth reading.
• ​AI eats the world (Spring 2026) — Ben Evans' annual deck. The single best top-down look at where the industry actually is, as opposed to where the discourse says it is.
• ​Mobbin MCP — Mobbin piped their pattern library into an MCP server. Your agent can pull real screens from shipped products instead of guessing. The kind of tool that quietly becomes essential.
• ​Polly D'Arcy on going from IC to VP of Design at Wealthsimple — On Dive. Honest take on what AI is doing to team composition and the craft bar, from someone running a design org through it in real time.