Kim Maida wants every agent tool call to mint a fresh, scoped OAuth token
AI Engineer · youtube · 2026-07-21
Kim Maida's talk proposes an old but effective fix for overprivileged agents: OAuth token exchange with short-lived, per-action credentials.
- The motivating incident is blunt: an incident agent followed a ticket, dropped the production Postgres database, failed to verify that the backup ran, and escalated it for the morning.
- The root cause was a single long-lived API key that could do everything, whether a human was watching or not.
- Their design mints a fresh token for every tool call, scopes it to just that action, and checks policy before the credential exists.
- That means dangerous actions like dropping a database are blocked before they can even be authorized, and the same pattern works across CLI agents, MCP servers, and OAuth providers.
More from coding & agent
- Soft Clamp cuts tool-call overuse in multi-teacher distillation, from 13.7% to 9.0% — antgroup · 2026-07-21
- Agent harness memory loss and compaction are still a major usability problem — adityaag · 2026-07-21
- SpecJudge runs locally on Ollama to pick the right-sized AI model for your project — jokiruiz · 2026-07-21
- A developer maps out six design rules for CLIs that humans and AI agents can both use — yujiezha · 2026-07-21
- A coding-agent skill that forces ADHD-friendly, answer-first output — ayghri · 2026-07-21
- A set of agent skills for CAD, robotics, and hardware design — earthtojake · 2026-07-21