Agentic AI adoption

I work agentically every day — Claude Code and Cursor as my primary environment, on a live production system. Most recently I built a CRM from scratch this way, replacing an outdated system that had been stitched together across a couple of Microsoft tools.

I slot into a team as a consultant with 10+ years in IT, automation and processes — now with agentic development on top. I help developers actually fold these tools into how they work: where the methods fit, what review discipline they need, and how to get real leverage instead of novelty.

My main worry is too much vibe in the vibe coding. Say an end user shouldn't see certain confidential data — if a feature was just vibed with no regard for that, the access rule was never built, and the wrong people end up seeing it. The technical judgement matters more once an AI is writing the code, not less.