🛠️Willison: Lines of Code Do Measure Agent Productivity
TL;DR
Simon Willison argues lines of code is a fair productivity signal with coding agents, because the pre-agent ceiling was real. A strong human day was 200 debugged lines and 50 to 60 was typical.
Simon Willison argues lines of code is a fair productivity signal with coding agents, because the pre-agent ceiling was real. A strong human day was 200 debugged lines and 50 to 60 was typical. Agents push that toward a thousand, but the new limit is cognitive capacity to review it.
Key Points
Pre-agent baseline: 200 lines of debugged production code is an excellent day, 50-60 is typical
Agents can reach roughly 1,000 lines a day at equivalent quality, but only with senior-level skill
Teams still need multiple engineers to load-balance review capacity, not typing throughput
Warns coding agents erode conceptual integrity, the Mythical Man-Month idea of software without surprises
From an August 2026 Talking Postgres episode with Claire Giordano
Why It Matters
The bottleneck moved from writing code to holding it in your head, which argues for smaller diffs and tighter scope rather than longer agent runs.
Quick Facts
Frequently Asked Questions
Why does this matter?
The bottleneck moved from writing code to holding it in your head, which argues for smaller diffs and tighter scope rather than longer agent runs.
What happened?
Simon Willison argues lines of code is a fair productivity signal with coding agents, because the pre-agent ceiling was real. A strong human day was 200 debugged lines and 50 to 60 was typical.
Comments
Be the first to comment
Enjoyed this article?
Get it daily. 7am. Free. Reads in 5 minutes.
Join 3,249 builders reading daily.