LLMOpsThe LLMOps lifecycle · MLflow 3 · FastAPI · Docker · AWS/Azure
Dashboard
0%
1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
41 small wins to finish your pathNext lesson →

LangGraph and AutoGen Studios

LangGraph Studio and AutoGen Studio are visual tools for building and debugging agentic apps: you see the agent's graph of nodes, step through a run, and inspect the state at each hop rather than reading it from logs.

Last updated: 15 Oct, 2026 · LLMOps

Tracing in Tracing a run with LangSmith recorded a chain after it ran. The studios work on the harder case: an agent that loops, branches and calls tools, where a flat trace is hard to follow. The Detailed LLMOps Project Lifecycle video names both while listing monitoring and debugging tools, and calls LangGraph Studio probably the best studio for an agentic application. They are named here, not run, because each needs its own install and a running agent to attach to.

What each studio gives you

  • LangGraph Studio draws the agent as the graph of nodes and edges you defined, runs it, and lets you watch the state move from node to node and step through or replay a run.
  • AutoGen Studio is the visual front end for AutoGen's multi-agent conversations: you compose agents and their tools in a UI and watch them talk, so you can debug who said what without writing the orchestration by hand.
  • Both target the same problem: an agent's control flow is a graph, and a graph is far easier to debug as a picture than as a stream of log lines.
Two panels. On the left a trace tool, used after the run to debug or monitor: three steps drawn as a staircase of nested boxes, a tree you read because it already happened. On the right a studio, used while building to see the flow work: an agent graph of circular nodes, plan branching to tool and think and both leading on to answer, with a dashed edge looping between tool and think, a graph you step through because it loops and branches.

The studios next to a trace tool

A trace tool (LangSmith)The studios
ShowsA recorded run as a treeThe live agent graph you can step through
Best forChains and single requestsLooping, branching agentic apps
When you reach for itAfter a run, to debug or monitorWhile building, to see the flow work

Where you use the studios

  • Building a LangGraph agent, where seeing the node graph makes a wrong edge or a stuck loop obvious.
  • An AutoGen multi-agent setup, where the bug is in which agent spoke when, not in any one model call.
  • Explaining an agent to a team, where the graph is the clearest way to show what the app does.
Watch out. A studio is a development and debugging surface, not a production monitor. It helps you see one agent run while you build; the cost, latency and error rates of live traffic still belong in a trace-and-metrics tool like LangSmith or Langfuse.
Try it yourself
  • Look up how LangGraph Studio attaches to a graph you have built and what it needs installed to run.
  • Compare AutoGen Studio's UI to writing the same multi-agent setup in code, and note what the picture makes easier.
  • For an agent you know, sketch its node graph on paper and mark where you would most want to pause a run.
PreviousTracing a run with LangSmith

You understood something today that you didn't yesterday.