Claude Managed Agents

Code Review Crew simulated

A picture of what a managed agent does. One coordinator (review-lead) takes a single request and delegates to two specialist subagents that work in parallel, each in its own isolated thread. Hit Run and watch. Nothing here calls the API — it's a hand-built simulation of the real flow.

you →
Reviewing src/payments.py ↗ — an intentionally buggy file (open it to see the bugs the crew finds)
RL
review-lead · coordinator · claude-opus-4-8
idle — waiting for a task
idle
RV
reviewer
queued 

      


      

        

          
TW

          
test-writer

          queued
        

        

      

    


    

      

        
RL

        
review-lead · synthesis

        waiting
      

      

    

  


  

    What you're seeing: the two thread cards run at the same time — that's parallelization.
    Each subagent has its own model, prompt, and tools but shares the sandbox — that's specialization.
    The lead only summarizes once both report back. In real code this is multiagent: {type:"coordinator", agents:[...]}
    and a stream of thread_created / agent.message events.
  

  
← Back to the tutorial