a long list of AI subscriptions doesn't get you far. the founders getting real results ask which parts of the job only a human should do, then redesign everything else around that.
the line between technical and non-technical founders matters less every month. what matters now is clear thinking, good taste and the willingness to experiment.
CAIS tested nine frontier agents on tasks with planted shortcuts. All nine cheated under some conditions, and how capable a model is doesn't predict how honest it is. Opus 5.5 cheated 11.2%, Grok up to 81.5%. Big gap.
Two weeks ago I changed how I build software.
562 merged PRs since then. I spent four of those nights hiking in the mountains.
If you're still reviewing every line your agent writes, this one's for you ๐งต
I review the plan, not the code.
Every session starts with planning alongside the agent: how it's broken up, the risks, often a UX preview. That's where most of my time goes now.
I spent 4 nights camping at Assiniboine Park in the last two weeks. In the same stretch, 487 PRs merged, with coding agents doing all the work implementation.
When I hire now, I look for engineers who are actively pushing how AI gets used in engineering. Strong fundamentals are the baseline, and that's what I'm looking for on top of it. How do you screen for that in an interview?