Anthropic Publishes Guide to Always-On Agent Architecture
Start here
Anthropic published a technical guide describing how to build always-on agents around five core primitives: an event inbox, timers, memory, permissions, and an outbox. The post says recent model improvements have made such agents more practical, citing Opus 5.5 costing 78% less per solved task than Opus 5 on one long-horizon coding benchmark, compared with 40% less on typical workloads. The guide also addresses the risks of handling permissions when no one is watching, covering credential proxies, auto-mode classifiers, and deferred tool calls that pause a run until the user approves. It does not report measured results from deployed always-on agents.
Generated from reports · updated 1 h ago
Report timeline
Follow the reports to see the event from different sides.
- Claude BlogPicksAnthropic explains the architecture behind always-on agents
Anthropic details the five core primitives behind always-on agents: an event inbox, timers, memory, permissions, and an outbox. The post says recent model gains made these agents more practical, citing Opus 5.5 costing 78% less per solved task than Opus 5 on one long-horizon coding benchmark versus 40% less on typical workloads. It also covers risks in unattended permission handling, including credential proxies, auto mode classifiers, and deferred tool calls that pause a run for user approval.
Heat for this event
Not enough continuous observations yet to draw a trend.