Diar Shakimov@diarshakimov1i asked Steve Wozniak, the co-founder of Apple, about coding in the age of AI.
Woz became a legend by obsessing over every tiny detail. Amjad Masad made me question whether that same instinct is holding engineers back with AI.
what if the next great engineer needs both?


i asked Steve Wozniak, the co-founder of Apple, about coding in the age of AI.
Woz became a legend by obsessing over every tiny detail. Amjad Masad made me question whether that same instinct is holding engineers back with AI.
what if the next great engineer needs both?
1/
Woz told a story about cutting a circuit board from 78 chips to 76. He'd wake up thinking about those two chips.
He's said the point was never cheaper manufacturing. He wanted to prove he was clever.
2/
Amjad said we should have at least 20 agents running. The opposite obsession.
Yesterday I wrote about asking @amasad why someone 3 months into vibe coding might beat someone 3 years in.
My takeaway: some engineers have spent 3 years getting better at asking AI to do less.
3/
That caution isn't irrational.
A lot of us watched early agents fail quietly. Tests passed while the real flow was broken. Answers cited functions that didn't exist.
So we made the tasks smaller, the prompts longer, the rules tighter. Our trust shrank. So did our ambition.
4/
Meanwhile a newcomer asks an agent for something ridiculous.
Most attempts fail. But the failures are cheap and private, and the few that work reset what they think is possible.
The expert's ceiling was set by a model from two years ago. The newcomer's was set this morning.
5/
I don't think Woz's obsession is wrong. I think it's pointed at the wrong thing.
Point it at the check instead of the code. A test that runs the real user path. An eval written before the change, so the agent can't grade its own homework.
6/
Permissions scoped so a bad run can't hurt anything. A source behind every claim. An alarm for the silent failure, not just the loud one.
Then give the agent a problem big enough to surprise you.
7/
This matters more for a team than for one person.
When an engineer gets burned, the lesson stays in their head. The next engineer and the next agent repeat the mistake, or inherit the fear without the reason.
Write down the failure, the fix and the check that proves it.
8/
My bet: Woz-level rigor, Amjad-level ambition. Be paranoid about the outcome and ambitious about the attempt.
What did you stop asking AI to do after it burned you once?
yesterday's thread on Amjad, if you missed it: x.com/diarshakimov1/st…