Developers debate how to test enterprise AI agents before real customers touch them
AstroAquatica · reddit · 2026-11-14
A developer building AI apps/agents for companies asks Reddit how others test agents before deployment. His Claude-spun-up sandboxes feel too clean: not messy like real companies, and since Claude helped build the system, it crafts scenarios it's guaranteed to pass. He's weighing mocked APIs, hand-filled test accounts, or cautious live rollout, highlighting a common pain point in agent eval engineering.
More from coding & agent
- WordPress UTM QA tip: accept cookies and submit forms without refreshing — HaktanSuren · 2026-11-14
- Using Claude to Visualize Ideas in Real Time During Team Calls — Flomerboy · 2026-11-14
- Vercel Token Pass doubles value: $150 in open-weight model tokens for $30/month — cramforce · 2026-11-14
- Dev moves nearly all code changes to subthreads, parent thread just orchestrates — zeeg · 2026-11-14
- Code-generated 3D Where's Waldo with 10,000+ animated people playable in browser — CSProfKGD · 2026-11-14
- Coding agent Junior goes macro-aware, sharing context across parallel sessions — zeeg · 2026-11-14