Anthropic's guide to always-on agents: sleep all day, wake for short runs, not 24/7
udmrzn · x · 2026-11-15
Anthropic published an engineering guide on building always-on agents, correcting a key misconception: they don't run 24/7 but sleep most of the day, wake for short runs, save state, and sleep again.
Key components distilled from the thread:
- Event Inbox: put a database table between webhooks and the agent; acknowledge webhooks fast, run one active worker per user to avoid collisions, and batch rapid messages.
- Timers: store in a separate table, split into one-off follow-ups and recurring schedules; a background poller drops due timers into the inbox.
- Lean Memory + Dreaming: store only what can't be re-derived (user instructions, notes, past messages, task progress); load a small index at wakeup; run periodic "dreaming" passes to purge stale facts.
- Unattended Permissions: never block waiting for user approval (truncated in source).
Practical reference for building memoryful, timer-driven autonomous agents.
More from coding & agent
- Database Architect: a Claude Code agent that designs schemas before you code — adnan_hashmi · 2026-11-15
- Teknium says his Hermes agents are now far more effective than Claude Code — Teknium · 2026-11-15
- Adobe hands CEO role to Anil Chakravarthy, previews local MCP for Photoshop, Illustrator, Premiere — shashib · 2026-11-15
- Running fully local AI to build a prompt-to-music-video pipeline: a deep walkthrough — julianharris · 2026-11-15
- CodeClimb: a coding agent whose reviewer keeps or reverts GraphRAG tuning experiments — blaizedsouza · 2026-11-15
- Dev ships first Steam game after 15 years, built on custom Rust engine with Claude — schiehll · 2026-11-15