Sentry CEO: trust agents by code location — Sentry changes get full human review
zeeg · x · 2026-11-22
Sentry CEO zeeg discusses with aidenybai where AI coding agents can run unsupervised ("yolo"). His take: it's about what the code is and where it lives — changes to Sentry itself always get human review; internal services with no customer data or uptime impact can be judged by how annoying a mistake would be. aidenybai admits recent yoloing went badly, and notes proper guardrails require making the mistake first. zeeg adds that in many scenarios no level of agent-generated trust is enough.
More from coding & agent
- Adding WebMCP to a Static Hugo Site: One Tool Call Saves ~99K Tokens vs HTML — DanWahlin · 2026-11-22
- Sentry CEO: reading the code beats explaining a multi-day outage from a bad LLM migration — zeeg · 2026-11-22
- levelsio declares "MCP is dead": agents are smart enough to just use APIs — zeeg · 2026-11-22
- Formalizing 1,826 tax types across 195 countries in a single Lean equation — SnooPeripherals5313 · 2026-11-22
- Code Mode cuts MCP token usage 99.9%: Cloudflare drops from 1.17M tokens to 1,000 — jasonkneen · 2026-11-22
- A game studio 3D scene built entirely from 4,903 lines of code — techartist_ · 2026-11-22