Descriptions:
Boris Cherny, the creator of Claude Code, says you should delete your CLAUDE.md, your skills, and your hooks every few months and see what the model does without them. Anthropic cut 80% of Claude Code’s own system prompt for Opus 5 and says it got smarter.
He’s right, but only to an extent. There is a lot more nuance here than the advice suggests, and I ran my own benchmarks on a real codebase to confirm where it holds up and where it falls apart.
Some of your AI layer genuinely has expired and you should be aggressive about cutting it. Some of it has not, and deleting that part costs you in ways that are surprisingly hard to notice.
In this video I break down what Boris actually said, what my own testing showed, and how to tell which of your rules to delete and which are still doing real work.
~~~~~~~~~~~~~~~~~~~~~~~~~~
CodeRabbit – AI code review that walks you through a PR change by change:
https://coderabbit.link/ad-cole-medin-001
~~~~~~~~~~~~~~~~~~~~~~~~~~
– I just hosted the free AI-Native Transformation Workshop, showing you how to effectively adopt AI as a standard for your engineering team. Register here to get the recording:
https://dynamous.ai/ai-native-engineering-org
– The skill that runs this ablation on your own repo, with any coding agent:
https://github.com/coleam00/skills
– Boris Cherny: We Cut 80% of Claude Code’s Prompt (the full talk):
~~~~~~~~~~~~~~~~~~~~~~~~~~
0:00 Delete Your Entire AI Layer?
1:16 What Boris Cherny Actually Said
2:48 Ablation Is an Eval, Not Just Deleting
3:28 The Over-Specification Mistake
4:21 Why Ablation Gets Expensive
5:16 The Unlimited Token Budget Problem
6:43 Sponsor: CodeRabbit
8:36 The AI Layer Spectrum
10:29 Where I Agree With Boris and Where I Don’t
12:38 Testing Full Ablation on Archon
15:02 Architecture Held Up, Conventions Broke
17:57 The Ablation Skill (Any Coding Agent)
19:38 Outro
~~~~~~~~~~~~~~~~~~~~~~~~~~
Join me as I push the limits of what is possible with AI. I’ll be uploading videos weekly – at least every Wednesday at 7:00 PM CDT!
