Tag: #programming

I Trusted My Own Safety Documentation. Then I Read the Test Suite.

I Trusted My Own Safety Documentation. Then I Read the Test Suite.

My safety doc said a control was enforced at the tool layer. It was enforced by nothing, and a passing test in the same repository had contradicted the doc for 21 days. The documented-but-unenforced control as a defect class.

Context Is a Contaminant: Your AI Reviewer Should Know Less

Context Is a Contaminant: Your AI Reviewer Should Know Less

A reviewer that knows who wrote the code and why reviews it worse. Independence comes from withholding provenance, not adding context. Your AI reviewer should know less.

The Verification Gap: We Automated Code Generation and Forgot to Scale Review

The Verification Gap: We Automated Code Generation and Forgot to Scale Review

We scaled code generation and left review behind. Four weeks of org data: 90.7% of pull requests AI-assisted, and at least half merged with no human review. The dashboards say review caught up because they count bots as reviewers. The gap, and how to close it.

The Hard Problem Isn't Automation. It's Classification.

The Hard Problem Isn't Automation. It's Classification.

Automating a bug pipeline is the easy part. Deciding which tier each bug belongs to is the bottleneck, and it is where the real engineering lives.

Your AI Skills Have the Same Problem as Your CLAUDE.md. Here's the Data.

Your AI Skills Have the Same Problem as Your CLAUDE.md. Here's the Data.

A controlled ablation, 42 runs: a behavioral-payload skill added zero marginal defect detection over a no-guidance control. A skill whose payload is documentation inherits documentation's 0%.

I Gave an AI Agent Unrestricted Shell Access. Here's What Happened Next.

I Gave an AI Agent Unrestricted Shell Access. Here's What Happened Next.

Unrestricted shell access handed to an AI agent, then the defense-in-depth architecture that made it survivable: layered hooks, and the self-approval bypass patterns they had to close.

Your CLAUDE.md Rules Achieve 0% Compliance. Here's the Data.

Your CLAUDE.md Rules Achieve 0% Compliance. Here's the Data.

A controlled ablation across nine tasks: a fully-specified CLAUDE.md changed agent behavior 0% of the time, identical to having no rules at all. The data, and why documentation cannot govern behavior.