AutoGen is a framework providing unified orchestration of conversations between multiple LLM-driven agents. Where LangGraph connects agents as a graph and CrewAI assembles them as a role-carrying team, AutoGen’s organising idea in the account given here is the conversation itself: agents are given roles and then talk to one another until a result is reached.
Capabilities
The source’s worked example of role separation is a travel-planning arrangement in which a
planner_agent produces an initial plan, a language_agent supplies language advice, and a
travel_summary_agent consolidates the several perspectives into a final plan. Of these the source
maps the first to the Planner role of a LLM-Based Multi-Agent System and the second
to the Reviewer role; it illustrates the Orchestrator role from a different framework rather than
from this example.
Its documented answer to disagreement between agents is a debate pattern, in which agents exchange positions over several rounds and correct one another. Solver agents propose answers and an Aggregator agent collects them, settling the result by majority vote over the solvers’ outputs from the final round. The source presents this as one of three arbitration strategies available to multi-agent systems generally, the others being confidence-weighted voting and trust-value routing; majority voting is the one it describes AutoGen as documenting.
For conversation structure the source names a RoundRobin mode, and for traceability it notes that callback functions can be used in AutoGen code to emit each agent’s decision detail into a unified log, so that a run’s reasoning can be correlated afterwards.
Adoption & Ecosystem
The source records that Microsoft merged AutoGen with Semantic Kernel into Microsoft Agent Framework in 2025, ending a period in which the company maintained two parallel multi-agent frameworks. It reads that consolidation as a signal that the field has moved from competing on features to converging on a few projects teams are willing to depend on, rather than migrating repeatedly between experimental ones.
Related Terms
- Microsoft Agent Framework — what it and Semantic Kernel were merged into
- LLM-Based Multi-Agent System — the arrangement this framework coordinates
- LangGraph — the graph-structured alternative the same source compares it with
- CrewAI — the role-based alternative the same source compares it with