How Much Inbox Access Should an Agent Get? Debating Fine-Grained Scopes
yenkel · x · 2026-11-19
A technical thread on permission granularity for AI agents touching user data. @yenkel argues the email API shouldn't treat "all emails" as the only scope: if a user asks to read one email, you shouldn't issue credentials that can read every email. Broad requests like "summarize my inbox" can stay broad, but APIs need finer-grained scopes.
@NattyChadwick asks whether the request sets the scope and policy decides issuance, and how a request whose scope is inherently every email gets narrowed — by time window or purely by policy. @yenkel replies that anything the API doesn't support needs a proxy policy layer.
Related event: How Much Inbox Access Should an Agent Get? Debating Fine-Grained Scopes(2 posts)→
More from coding & agent
- Flow Engineering Brings Agentic AI to Hardware Design; Rivian Engineers Grew 40 to 1,500 in 4 Months — PTrubey · 2026-11-19
- Linq raises $90M Series B led by a16z, has processed 1B+ agent messages — a16z · 2026-11-19
- Google Labs launches Stitch Loop, an always-on product agent for post-launch improvement — GoogleLabs · 2026-11-19
- Email APIs need finer scopes for agents: treat inbox as a tag, read by range — yenkel · 2026-11-19
- Same models, more bugs: Xint's harness finds 5 critical nginx bugs, 3 never found before — dyn___ · 2026-11-19
- lineup: Open-Source CLI That Catches Coding Agents Making Sneaky Edits — Several-Coyote3534 · 2026-11-19