"Use AI for everything or you're finished" versus "AI is just autocomplete for cheaters" — both positions are tiring, and both are wrong. This talk is the pragmatic middle: how a real engineering team produces code with AI day to day. I'll walk through everyday AI-assisted coding (where it earns its keep, and where we keep humans firmly driving), generative tooling we built (plain English to live BigQuery charts that replaced one-off dashboards), and an agentic pipeline that triages Sentry errors and drafts a fix but deliberately stops at a pull request, never a merge. The throughline is AI proposes, humans dispose — but the part most talks skip is the 'without deskilling the team' bit: avoiding review fatigue, over-trust, and the morale hit of feeling replaceable. You'll leave with a practical way to adopt AI on your team without the FOMO and without eroding the craft or the people, plus a checklist for where to draw your own automate-versus-gate line.