01What Claude Managed Agents Updates Are and How They Work

Claude managed agents updates are four new features from Anthropic that make AI agents work like a self-improving team. Announced at the 'Code with Claude' event on May 6th, these updates include dreaming, outcomes, multi-agent orchestration, and webhooks.

Dreaming lets an agent review its past sessions overnight, clean up messy memory, and write better notes for next time. Outcomes adds a separate grader agent that checks work against a checklist and orders rewrites until it passes. Multi-agent orchestration turns one big job into parallel tasks handled by specialist agents.

Webhooks let another app ping you when the job is done. Together, they turn Claude from a forgetful assistant into a team that learns.

02Main Features of the Claude Managed Agents Updates

The headline feature is dreaming. It runs as a scheduled background process between sessions. It reads past work, finds patterns, cleans up duplicate or outdated memory, and writes new notes for future tasks.

  • Outcomes: You write a checklist of a good result. A separate grader agent reviews the final work against that list. If it fails, the main agent tries again.
  • Multi-agent orchestration: A lead agent breaks a big job into small pieces. Specialist agents, each with its own model, prompt, and tools, work in parallel and share files and notes.
  • Webhooks: Kick off an agent, walk away, and get a notification in Slack, email, or your own app when it finishes.

Dreaming is in research preview. The other three are in public beta. Developers can request access today.

03How to Set Up or Get Started with Claude Managed Agents

Getting started is easier than you think. You do not need to be a coder. You need a clear job, a checklist for what good looks like, and a starting prompt.

  1. Go to the Claude managed agents documentation on Anthropic's site.
  2. Request access for the features you want. Outcomes, multi-agent orchestration, and webhooks are in public beta. Dreaming is in research preview.
  3. Pick one simple task you do every week. For example, writing a short report or answering customer emails.
  4. Write a clear rubric or checklist that defines what a good result looks like.
  5. Set up your first agent with a starting prompt that describes the job.
  6. For outcomes, add a second agent as the grader. Give it the checklist and have it check the final work.
  7. Test the agent on a few runs. Watch how it improves with dreaming enabled.

04How to Use the New Features Step by Step

Here is how the features work together in practice, based on what Anthropic showed.

Start with one agent doing a task. After it finishes, it writes notes to memory. Over time, that memory gets messy.

Enable dreaming so it runs overnight and cleans up those notes.

  • For outcomes, write a checklist of what good looks like. Then let the grader agent review each piece of work. If it fails, the main agent rewrites it until it passes.
  • For multi-agent orchestration, set up a lead agent that splits the job. Each specialist gets its own prompt and tools. They share files and notes as they work.
  • For webhooks, connect the agent to Slack, email, or your app. Kick off the job and walk away. You will get a notification when it is done.

In the space mission example, they ran a simulation, pressed one button to trigger dreaming overnight, and the next morning the new run was much better.

05Results and Comparisons from the Episode

The numbers in this episode are striking. Harvey, a legal AI company, turned on dreaming for their document handling agents. Their task completion rates went up around six times.

That improvement came from the agents remembering small things like file format quirks and tool-specific tricks. It was not a new model. It was simply the agent carrying lessons across sessions.

Another example is WiseDocs, a medical document review company. They built a quality check agent using outcomes. WiseDocs cut its document review time by 50%.

Netflix also got a mention. They used Claude agents to process logs they could never get through manually.

06Tips, Limits, and Mistakes to Avoid with These Updates

Here are some practical tips from the episode.

  • Start with one task. Do not try to automate everything at once.
  • Outcomes is the easiest feature to try first. Pick a weekly task, write a rubric, and run it.
  • Remember that dreaming runs between sessions. Give it time to work overnight.
  • The grader in outcomes sees only the final result, not the process. That avoids bias.

Be aware of the limits. Dreaming is in research preview, so it may not be ready for production use. The other features are in beta, so expect changes.

A common mistake is ignoring these updates and doing nothing. In six months, competitors who use them will be far ahead.

07Who It Is For and Why It Matters

These updates matter for small business owners, marketers, coaches, and service providers. You do not need to be a coder to use them.

If you use Claude like a smart assistant today—ask a question, get an answer—this update lets you use Claude like a team. You give it a goal. The team runs in the background, checks each other's work, learns every week, and pings you when done.

The big shift is from one chat at a time to a team of agents running your repeat tasks while you sleep. That is the real story.