Pragmatic Testing for Agents
05 October, 2026My agents kept writing tests that restate the code. I gave them a bar a test has to clear, deleted fourteen thousand lines, and used coverage and mutants to check nothing that mattered went with them.
7 posts found. View all tags
My agents kept writing tests that restate the code. I gave them a bar a test has to clear, deleted fourteen thousand lines, and used coverage and mutants to check nothing that mattered went with them.
How we made Xcode and XcodeBuildMCP reuse build outputs, proved that PATH differences trigger linking, and decided the last few seconds were not worth a workaround.
My agent rewrote a module and told me the result was better. Here is how I made it prove that with numbers instead of adjectives.
How a simple prompt from a Matt Pocock video turned into a structured pair of AI agent skills that interrogate product ideas and technical designs the way a senior colleague would.
I built a package that plays Dota 2 hero voice lines as notifications for Claude Code, OpenCode, and Pi. Here's why, how, and what I learned about hook systems across three different AI coding tools.
How Tuist's binary caching, code-based manifests, and deterministic setup make the git worktrees workflow practical for parallel AI agent development in iOS.
Why AI agents demand we abandon medieval craftsmanship for industrial software development - and why that's a good thing.