When AI Costs More Than the Engineer | Exec Engineering #200
Also: why AI asks for more discipline, not less, why killing a project is a leader's hardest call, and the All Hands half your team is tuning out.
Hey👋
Thanks for reading Exec Engineering, a weekly digest for the busy tech executive.
I hope this edition brings you value.
The Digest
When AI Costs More Than the Engineer (Tomasz Tunguz)
Anthropic spends 2.3 times its whole payroll on raw compute. The number worth watching is what you pay in AI for each engineer. Today the top 1% of firms spend about $89k per engineer, close to 40% of a senior salary, while the middle spends $137 a year. Tunguz maps three ways that gap could close by 2029, and past a point the AI bill stops being optional. Worth a read to see which of those futures you are paying for.
Killing a project is every engineering leader’s hardest call (Alessandro Bahgat / LeadDev)
Your job is to fight for your team, which is not the same as fighting for every project they are on. Bahgat defended one at Google for years before it got cancelled, and what he felt after was guilt, for keeping good people on a dead end. He now runs new bets past a few plain questions first. When you cannot say why us, or what you give up to do it, you have your answer.
AI Demands More Engineering Discipline. Not Less (Charity Majors)
A model can now write fine code for almost nothing, so the hard part moves to reviewing it and owning what it hands you. Charity Majors argues this asks for more care than before. A sharp reply to anyone who thinks cheap code lets a team get sloppy.
So You Want to Fix Your All Hands (Michael Lopp)
Half your team quietly thinks the All Hands wastes their time, and the real reason is easy to miss: it has slowly become a meeting about the leader, not the room. The fix flips that: curiosity keeps people awake, so you build the hour around what they actually want to know. Steal one rule today and cap it at 90 minutes. How to fill the rest is the part worth reading.
The 5 Impacts of the Owner-Operator Model on On-call, Operations, and Incentives (David Anderson)
Every year Amazon froze all code over the holidays, and outages dropped by about ten times. That fits what the data shows, since most outages start with a recent change. So the surest fix for quality is to put whoever wrote the code on call for it. A 3am page from your own bug teaches more than any lecture can.
Is AI unbundling expertise? (Matt Kamelman / Thoughtworks)
The clearest read yet on Anthropic’s big study of 400,000 coding sessions, and it lands somewhere odd. What the tool rewards is not how much you know. It is how well you can hand your thinking to a machine. The accountant who knows the exact rules a script must follow beats the junior dev who cannot say what he wants. So the thing to grow in your people is the habit of spelling out how they think.
Default to Lazy (Aviv Ben-Yosef)
Larry Wall called laziness a good trait in an engineer. Aviv Ben-Yosef makes the same case for whole teams. People who never slow down fill the gap with busy work and tie their worth to raw output. Reward the ones who quietly solve a thing once over the ones who love to play hero.
When Impressive Performance Gains Do Not Matter (Colin Breck)
A query cut from 10 minutes to 1 sounds huge, and Colin Breck explains why your users will not feel it. Past about 10 seconds, people go check their phone, so 10 minutes and 1 minute land the same in their day. A good gut check before a 10x number goes on a slide. Does it cross the line where someone would notice?
Dialog
I recently talked to Manuel Drews, who runs engineering at Native Instruments, about the moment a leader’s hard-won instincts quietly stop working.
His case is that the playbook you trusted at one size quietly breaks as the org changes, and rarely with a bang. It creeps in as plain friction, easy to blame on the wrong thing.
Take a team that looks broken. His read is that it is usually just badly shaped, and the giveaway is that the smart, motivated people inside it keep struggling anyway.
In the full conversation he gets into what to do about a team like that, and where he draws the line on the rush to ship faster with AI.
More reads
Homo Agenticus (Strange Loop Canon)
I Shipped a Facebook Feature So Fast Sheryl Sandberg Called an Emergency Meeting to Stop Me (Michael Novati)
The Linguistics of Programming (in|retrospect (Allen Cheung))
Beware of Productivity Paradoxes (Cal Newport)
TBM 427: The Bottleneck Strikes Again (The Beautiful Mess (John Cutler))
This digest is backed by Gemography.
Augment your engineering team with remote high-ownership engineers from Africa and LatAm, on demand. Meet precise matches for your stack and culture under 48h.
About Exec Engineering
I’m Yassine 👋 I spend a big chunk of my time digging into engineering management and talent acquisition, especially where the two overlap. I share the most interesting resources I come across in this newsletter, all curated by hand.