Thoughts on Slowing the F*** Down — Simon Willison
URL: https://simonwillison.net/2026/Mar/25/thoughts-on-slowing-the-fuck-down/
Author: Simon Willison
Published: March 25, 2026
Summary
Simon Willison’s link post endorsing Mario Zechner’s critique of agentic coding speed. The core argument: agents let you generate code faster than you can understand it. The bottleneck shifts from typing to comprehension — but without the natural speed limit humans provided, mistakes compound until the codebase becomes unmanageable.
Key Claims
- Human as bottleneck was a feature: A human “cannot shit out 20,000 lines of code in a few hours.” This pace matched human ability to reason about the codebase.
- Agent mistakes compound differently: Many small “booboos” accumulate at machine speed. Each individually seems harmless; together they form “a monster of a codebase.”
- Cognitive debt is real: Changes that would normally be considered over weeks land in hours. The codebase evolves outside the developer’s ability to reason about it.
- Mario Zechner’s recommendations: Give yourself time to think. Set limits on how much code agents generate per day (calibrated to your review capacity). Write architecture, API design, and core structure by hand.
- Willison’s nuance: Writing by hand isn’t necessarily the right fix — but discipline and balance between speed and mental thoroughness is necessary now that typing is no longer the bottleneck.
- The new bottleneck: Human understanding and review capacity, not typing speed.
Context
Mario Zechner created the Pi agent framework used by OpenClaw — giving his critique practitioner credibility. This is not abstract concern; it’s from someone who built serious agentic infrastructure.