3 Things Worth Reading #002
Self-Taught Engineer weekly reading edition - from the beach
I read a lot of engineering newsletters so you don’t have to. Most weeks, three things survive but this week I spent some extra time reading my kindle with a beer in hand. This week there all point at the same idea from different angles: fundamentals are getting more valuable, not less.
1. Napkin math is a superpower
Pushing software engineering limits with “napkin math"
The Pragmatic Engineer, with Turbopuffer cofounder Simon Eskildsen
Simon Eskildsen builds vector databases from first principles. His core habit is simple: before you build anything, estimate what the system should cost from the raw numbers disk latency, memory bandwidth, dollars per GB. If your design is 10x off the napkin, the design is wrong. I remember reading an interview that Simon did years ago about how he learned and digging into his well written blog and how he used anki cards which is a lesson in being a self taught learning machine.
Why this matters: napkin math is not a CS-degree skill. It’s a skill you learn it by caring about a real problem and working backwards. Sometimes called the Fermi Estimate. The story goes that Enrico Fermi estimated the yield of the first atomic bomb by dropping scraps of paper and watching how far the blast pushed them off by less than a factor of two.
You need the estimate, and the honesty to compare it with reality my friend Nick Zervoudis wrote a bit about using it for measuring data products
2. AI writes half the code now. Judgment hasn’t been automated, yet?
The State of AI Impact in Engineering
Refactoring (Luca Rossi + Justin Reock), on data from 500+ teams
The numbers: over 50% of code is now AI-generated, and median throughput is up 37% in a year. Great, right?
Pull request sizes nearly doubled. Developer experience metrics declined. The gains seem to concentrate in small orgs; big enterprises are stalling across the board at the moment.
Tools are getting faster. The scarce thing is still judgment I’m interested to see where this is headed and my linkedin dm’s are full for all the tools.
3. The best CS lecture you never took is free
A Beginner’s Guide to Clocks, Causality, and Ordering in Distributed Systems *paid
ByteByteGo
Why is “what time is it” one of the hardest questions in distributed systems? This piece explores clocks, causality, and the ordering issues behind every “why is this event out of order?”
When people ask me how to break into data engineering without a CS background, this is what I point to. Not a bootcamp. Not this week’s latest LLM. The principles we discuss are true now, just as they have been for decades. They will still be relevant when today and tomorrow.
Learn principles. The tools will change under you anyway and almost everything that ByteByteGo is worth investing in.
---
Bonus: go straight to the data
The State of AI Impact in Engineering: Q2 2026
DX the primary report behind The State of AI Impact in Engineering
DX’s quarterly report is where the 500+ team numbers come from AI-generated code share, throughput, PR size, developer experience, all broken out by org size. DX’s researchers are the same folks whose work keeps showing up on The Pragmatic Engineer, and this is the rare AI-impact dataset that reports the uncomfortable numbers alongside the flattering ones.
---
The thread through all three: AI is commoditizing the output of engineering code, queries, pipelines. It is not commoditizing estimation, judgment, or fundamentals. Those you still have to earn the old way. Which is the only way you’ve ever earned anything.
What did you read this week that survived the filter? Hit reply I read everything.
I also write/podcast about raising four kids with my neuroscientist wife over at Little Big Humans