Managed agent workflows enter beta

Claude Managed Agents dynamic workflows have entered public beta, according to Addy Osmani’s post. The setup uses a lead agent to create a plan, then runs work across agents in phases and combines their results. Osmani says it can orchestrate up to 1,000 agents per run.

The example in the post tested the approach on a codebase with 70 planted bugs. A single agent found 14, 15 and 27 bugs across three runs. The dynamic workflow found 66 of 70 in each of its three runs. That’s a result from one described test, not evidence that every business task will see the same improvement. The post also says these workflows can use a lot of tokens and recommends starting with a scoped task, then increasing complexity gradually.

For business owners, the bigger signal is that coordinating agent work is being offered as a managed workflow, not just as a way to ask one AI tool a question. That could matter for companies with work that breaks into stages, but the post gives no pricing or adoption figures, so it’s too early to judge the business case. Before trying it, decide what task to test and what boundaries it needs. Teams may also want to track how AI work is set up and used as they test this kind of workflow.