Rate limiting agent write verbs: counting rows enforces nothing, log what they did
Silver_Employ2617 · reddit · 2026-09-13
A developer running an AI-art site's MCP server recounts his rate-limiting mistake: counting current rows let agents loop like→unlike→like forever, always showing count 1 while spamming real notifications.
His fix and practices:
- Rate limits should track actions, not state — use an append-only action log (schema included)
- Unlike/unfollow should burn budget too, or undo becomes a free re-notify
- Caps apply only to agent tokens (e.g. like 60/h, comment 15/h); humans in browsers stay unlimited
- Naming each tool's cap in its description makes models back off instead of retrying into naked 429s
He notes 21 real agent tokens minted, zero used, and asks peers for real usage data on write verbs.
More from coding & agent
- Stripe offers one-shot agent wallet integration, already powering Grok Bot and Muse — jeff_weinstein · 2026-09-13
- From PM to GTM Engineering: You Don't Have to Write Code to Be Technical — brandon_galang · 2026-09-13
- Dev goes 'voice pilled' with Wispr Flow, dictating 15-minute prompts to his agent — fernandorojo · 2026-09-13
- Labeler Agreement Check: If Two Trained Labelers Can't Agree, the Rubric Is the Bug — blaizedsouza · 2026-09-13
- Learn Omarchy: interactive desktop course with 17 lessons and 101 activities for DHH's distro — DanWahlin · 2026-09-13
- Better than /goal: orchestrate coding subagents with a top model and integration criteria — gregmushen · 2026-09-13