Posts tagged “#claude code”

Last week an AI coding agent deleted a production database in 9 seconds. The thread documenting it and The Register's coverage have been read as a story about AI safety: a flagship model went rogue, the guardrails didn't hold, another company… (more)

A low-risk commit is fine on its own. Ten low-risk commits that haven't been pushed are not fine. The risk compounds. Each commit adds to a batch that will eventually land on a CI pipeline, and that batch has its own risk profile that no individual… (more)

An AI agent makes architectural decisions constantly. Add a dependency, change a build script, restructure a config file. Each choice is reasonable in isolation. None of them get written down. This is the knowledge management version of technical… (more)

An AI agent writes fluent prose. It also writes whatever you ask for. Tell it to add a hero section and it will produce "We're passionate about leveraging cutting-edge solutions." Tell it to write an FAQ answer and it will open with "Great… (more)

An AI agent has no visibility into accumulating work-in-progress. It works on the current prompt. Meanwhile, uncommitted changes grow, commits pile up without a push, changesets go unwritten, and a release PR sits open for days. The accumulation… (more)

In the previous post I described a pipeline where every push to main ends with a live preview of the production candidate, smoke-tested and ready for human review. The idea is that nothing ships without you seeing it running. There's a gap, though.… (more)