How much do you let your agents do unsupervised? Devs debate production approval boundaries
Weird_Impression_264 · reddit · 2026-11-19
Reddit user WeirdImpression264 asks where people draw the line for autonomous agents in production: reading data and answering questions feels safe to automate, but updating records, emailing customers, or touching money feels too risky to leave to the model.
The post canvases three common approaches — human approval gates before sensitive actions, hard limits enforced in code, or trusting the prompt and watching logs — and invites war stories about what worked and what went wrong. A typical practitioner discussion on agent permissioning in real deployments.
More from coding & agent
- I8-Project Uses AI Agents to Crack Long-Open Higher-Curvature Calculations in 11D Supergravity — burny_tech · 2026-11-19
- 10-year OSS veteran open-sources cribelo: AST + git history to triage 40-file agent PRs — NoMedium7071 · 2026-11-19
- Choosing Between Redis, Kafka, Elasticsearch and PostgreSQL: Start With the Workload — goyalshaliniuk · 2026-11-19
- Dev Celebrates Agents Rewriting Everything to Rust as TypeScript and Ruby 'Vanish' — smlpth · 2026-11-19
- Gemini agent unpromptedly trains an ML model on user's Spotify taste while fixing MCP — lyraxana · 2026-11-19
- SocialCrawl launches Discover API: one MCP endpoint to search Reddit, YouTube, X and more — dooddyman · 2026-11-19