From Semantic Kernel to Microsoft Agent Framework (After-Work Guide)

Semantic Kernel (SK) is Microsoft’s SDK for prompts, plugins, and planners. In 2026 the product story is convergence: new agent work should lean on Microsoft Agent Framework, with SK docs as context and migration fuel—not a forever “start from zero on SK only” path.

Read: Semantic Kernel overview and Agent Framework.

Who this is for

After-work Python learners who already met SK plugins/kernels, or who work near Azure OpenAI and need the current recommended agent entry. Absolute beginners can still follow—just treat Agent Framework as the default install target.

Overview: Top 10 AI agent tools for after-work Python.

Steps to start tonight

1. Skim the migration intent (10 minutes)

Open both Learn pages. Write one sentence in NOTES.md: “I will build new agents on Agent Framework; I open SK docs when I must port an old plugin.” That sentence is the whole angle of this post.

2. Create a venv and install what the Agent Framework quickstart lists

python -m venv .venv
source .venv/bin/activate
# install packages named in the current Agent Framework Python quickstart
# (names change—copy from learn.microsoft.com/agent-framework, not from memory)

If you still have an old semantic-kernel project, leave it in another folder. Do not mix half-migrated imports on night one.

3. Configure one Azure OpenAI or OpenAI credential

Use environment variables / dotnet user-secrets-style local secrets as the docs show. Never commit connection strings.

4. Run the shortest Agent Framework agent sample

Copy the minimal “chat agent + one tool” (or equivalent) from the Agent Framework docs. Invoke once with: “Return the square of 12 as an integer only.” Print the result and any tool trace the sample exposes.

5. Optional SK contrast (15 minutes)

If you have an old SK notebook, list which concepts map (plugins ≈ tools, kernel ≈ host). Do not rebuild the SK sample—just label the vocabulary so blog posts from 2024 still parse.

Tiny exercise (20–30 minutes)

Add a second tool add(a, b) (or rename per docs). Ask: “What is 7 plus 5?” Confirm the framework routed to your tool. If it did not, improve the tool description—do not jump back to a full SK planner demo.

Pitfalls

• Teaching SK as the forever greenfield default. The Hub angle is migration / Agent Framework first.
• Porting every SK planner sample on night one. One tool-calling agent is enough.
• Mixing SK and Agent Framework imports until something “kinda works.” Separate folders.
• Skipping Learn.com version notes. Microsoft samples move; pin the page you copied.
• Secrets in source. Same as always.

Why this fits after work

You get an honest map of Microsoft’s agent stack in one evening—without pretending SK tutorials from two years ago are still the front door.

Next step

When typed tools and host setup feel clear, deepen Python foundations with Intro to Python Programming so SDK samples stop feeling like ritual.