Anthropic Outlines Five Primitives for Always-On Agents
Understand the event
Anthropic outlined five primitives for building always-on agents: an event inbox, timers, memory, permissions, and an outbox. Together, they coordinate brief runs across sessions, resume work after user input, and manage tool calls made without a person present. The post describes options for implementing the components, including durable databases or object stores for memory and a scheduler or durable execution engine for timers. Dotey separately explained the same five components through an AI assistant, saying it can remain idle much of the time without calling a model or incurring costs.
Generated by AI from reports · Updated 2 days ago
Read next
More changes in today’s briefingGemini Connects to Google Wallet for Personal Finance InsightsDevelopments
2 developments
- 11/14 08:56 (UTC+8) · 1 reportsDotey Explains How Always-On AI Agents WorkBaoyu:How Always-On AI Agents Work: Five Components Behind the Assistant
- 11/13 16:00 (UTC+8) · 1 reportsAnthropic Explains the Architecture of Always-On AgentsClaude Blog:Building Always-On Agents: Five Primitives for Their Architecture
Report timeline
Follow the reports to explore different sides of the event.
- How Always-On AI Agents Work: Five Components Behind the Assistant
Baoyu explains an always-on AI assistant through five components: an inbox that wakes it for incoming tasks, timers for proactive actions, memory files, permissions, and an outbox that controls when it speaks. The post says the assistant spends much of its time idle without calling a model or incurring costs, and previews how the setup handles simultaneous tasks, memory limits, approval for important actions, and quiet operation.
- Claude BlogPicksBuilding Always-On Agents: Five Primitives for Their Architecture
Anthropic’s post explains how to build always-on agents from five primitives: an event inbox, timers, memory, permissions, and an outbox. It describes how these components coordinate brief runs across sessions, including resuming work after user input and managing unattended tool calls. The post also compares options such as storing memories in durable databases or object stores and using a scheduler or durable execution engine for timers.
Keep up with AI, one email a day
Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue
Event heat trend
- Current within comparable coverage
- 3
- Peak within comparable coverage
- 1011/14 10:00 (UTC+8)
- 24-hour change
- -49%
Trends compare the same continuously observed participants, which may cover fewer sources than current heat. Point or click to inspect hourly heat, or use left and right arrow keys.