Working practice for building with AI coding tools: where they hold up, where they fail, and what the failure modes cost in review time. Evaluations are run against real repositories rather than benchmarks, and re-run when the underlying models change, because a result from six months ago is not a result about the tool you are using today.
AI isn't replacing senior engineers, it's changing what the job is: the real shift from writing code to directing and supervising the agents that write it.
MCP servers aren't just another API layer. Here's what actually happens when a coding agent calls one, with a real setup, not the spec restated.
Vibe coding isn't inherently bad, unreviewed output shipped somewhere it matters is. Here's exactly what breaks when nobody checks it, and why.
We migrated the same 60-file API through Claude Code, Cursor, and GitHub Copilot. Here's what each one missed, and which one we'd trust with production code.
What to commit, what to gitignore, and why the CLAUDE.md, Cursor rules, or AGENTS.md file drifts out of date faster than anything else in the repo.