A Week of Major Moves
This was one of those weeks where you could feel the Claude ecosystem shifting into a higher gear. Anthropic launched Managed Agents into public beta, revealed more about its cybersecurity initiative Project Glasswing, and navigated through a rough patch of service outages. Let's walk through all of it.
Claude Managed Agents — Making AI Agent Deployment Accessible
On April 8, Anthropic announced the public beta of Claude Managed Agents, a fully managed infrastructure service for running autonomous Claude-powered agents.
Until now, putting AI agents into production has required building orchestration layers, handling scaling, monitoring, and failure recovery — all complex infrastructure work that has little to do with the agent's actual logic. Managed Agents takes care of all of this, letting developers focus purely on what their agents should do rather than how to keep them running.
The pricing model is straightforward: $0.08 per session hour on top of standard Claude API token costs. That puts agent deployment within reach for individual developers and small teams, not just well-resourced enterprises.
What stands out most is the built-in monitoring and autoscaling. Long-running workflows like data analysis pipelines, automated customer support, and document processing can now be deployed without worrying about infrastructure complexity. This is one of those changes that could significantly lower the barrier to building real-world AI agent applications.
Project Glasswing and Claude Mythos Preview
The other headline this week was the broader reveal of Project Glasswing and the limited release of Claude Mythos Preview.
Mythos Preview is a specialized model capable of identifying and exploiting software vulnerabilities with what Anthropic describes as "unprecedented accuracy." Rather than making it publicly available, Anthropic is restricting access to over 50 tech organizations — including AWS, Apple, and Cisco — along with more than $100 million in usage credits to help these companies find and fix vulnerabilities in their foundational systems.
The decision to keep such a powerful model gated is a meaningful statement about responsible AI deployment. By channeling this capability exclusively toward defensive cybersecurity, Anthropic is demonstrating that raw capability and careful stewardship can coexist. For users and developers in the Claude ecosystem, this kind of measured approach to powerful technology is reassuring.
Weekend Service Outages — April 6–8
The week started on a rough note, with significant Claude AI service disruptions over the weekend and into early Monday.
The first outage hit around 10:30 a.m. Eastern Time on April 6, with Downdetector registering approximately 2,700 reports within minutes. Another spike followed on April 7, and some users reported instability into April 8.
Service outages are an inevitable part of a rapidly growing platform, but they're a reminder of the importance of building resilience into any workflow that depends on Claude. If you're using the API in production, now is a good time to review your retry strategies, implement proper exponential backoff, and consider fallback options for critical paths.
Claude Code Keeps Evolving
Claude Code continued its steady cadence of improvements this week.
/powerup command gaining traction: Introduced in early April, the /powerup command provides interactive lessons that teach Claude Code features right inside the terminal. Covering context management, Hooks, MCP, sub-agents, and the /loop command, each lesson takes 3–10 minutes and includes animated demos. It's an elegant solution to the learning curve that keeps you in your flow state rather than bouncing to external documentation.
Expanded MCP result persistence: MCP tool results can now be retained up to 500K, which makes a real difference for workflows involving larger datasets.
Stability improvements: Fixes across resume functionality, plan mode, terminal handling, and Windows support. If you've been experiencing rough edges in your daily Claude Code usage, it's worth updating to the latest version.
Recommended Reads from This Week
Claude Lab published a range of articles this week. Here are some highlights worth checking out.
On the Claude Code front, we covered troubleshooting VS Code integration issues, using the /insights command to analyze your coding habits, and building cross-platform apps with Expo and React Native.
For API and SDK developers, we published a comprehensive guide to the new Managed Agents service, explored the Files API for document persistence and cost reduction, and dove into building AI REST APIs with Hono.js.
In Claude AI general topics, we analyzed the latest Anthropic IPO developments, explained the Claude Powerup features, and put together a guide for integrating Claude with Google Workspace.
What to Watch Next Week
A few things to keep on your radar heading into next week.
Sonnet 4.5 1M Context Deprecation Countdown: The April 30 deadline for the 1M token context window beta on Sonnet 4.5 and Sonnet 4 is approaching. If you haven't migrated to Sonnet 4.6 or Opus 4.6 yet, now is the time.
Managed Agents Real-World Reports: With the public beta just days old, we should start seeing community feedback and practical use cases emerge. Claude Lab is preparing implementation guides as well.
Service Stability Follow-Up: After this week's outages, it will be worth watching how Anthropic addresses reliability concerns.
Wrapping Up
The launch of Managed Agents feels like a turning point — AI agents moving from experimental curiosity to everyday infrastructure. Project Glasswing shows that Anthropic is thinking carefully about how its most powerful models should be used. And the service outages, while frustrating, are growing pains for a platform that's clearly scaling fast.
We'll keep tracking all the developments here at Claude Lab. Thanks for reading, and see you next week.