I’m a Business Analyst trying to decide if I should learn AutoGen or LangGraph. My tasks involve pulling data, creating charts, and then summarizing findings into a report. Which one is easier to maintain when the "state" of the analysis needs to persist over several days?
3 answers
For data-heavy Business Analysis, AutoGen has a slight edge due to its native handling of tool calls and data visualizations. If an agent needs to generate a Matplotlib chart, it just writes the code and shows it. Regarding persistence, while LangGraph is very "explicit" with its graphs, AutoGen uses a simpler session-based memory that can be backed by a SQL database. For tasks that span days, you just save the conversation thread and the generated files. It feels more like working with a digital intern who keeps a notebook. LangGraph is better if you have a very rigid, flowchart-like process, but for the messy, iterative nature of analysis, the conversational flow here is much more intuitive.
If the "intern" (agent) makes a mistake on day one, is it easy to go back and correct just that part without restarting the whole analysis?
I prefer the conversational style for reporting. It's much easier to explain to a client how the AI reached a conclusion.
Exactly, Elaine. The transparency of the agent "discussion" makes the audit trail for Business Analysis very clear.
That’s where the "Human-in-the-loop" feature comes in, Carlton. In AutoGen, you can set the human_input_mode to "ALWAYS" or "TERMINATE." This allows you to intercept the agent's work at any step. If you see the data pull was wrong on day one, you just type the correction in the chat, and the agent adjusts its path immediately. You don't have to rewire a graph; you just talk to the system.