Anthropic Launches Managed Agents API and Graduates Claude Cowork to GA
What Happened
Anthropic released Claude Managed Agents in public beta — a suite of composable APIs for building and deploying cloud-hosted agents at scale, handling sandboxed code execution, checkpointing, credential management, and end-to-end tracing. Simultaneously, Claude Cowork (the desktop computer-use agent) graduated from research preview to general availability with enterprise features including role-based access controls, group spend limits, and usage analytics. Early adopters include Notion, Asana, and Sentry. Managed Agents pricing is standard Claude API rates plus $0.08 per session-hour of active runtime.
My Take
This is Anthropic's real play, and it is not about models. Managed Agents is infrastructure-as-a-service for AI workflows — sandboxed execution, checkpointing, credential management — the boring, hard stuff that separates a demo from a deployment. The phrase from The Register captures the pitch perfectly: "run your business on autopilot." And here is the problem: most teams building with AI agents today are duct-taping together their own orchestration, burning weeks on infra that has nothing to do with their actual product. Anthropic just said "stop doing that, we will handle it." At $0.08 per session-hour on top of API costs, the pricing is aggressive enough to pull teams off homegrown solutions. For product engineers and founders building agent-powered workflows, this compresses the timeline from prototype to production dramatically. But the real question is trust. These agents "read files, run commands, browse the web, and execute code" with reduced human oversight. That is a lot of autonomy to hand over to a hosted service. The teams that win here will be the ones who understand what to delegate and what to keep human — judgment over speed, as always.
Read Original Source