Compare orchestration frameworks using the same acceptance tests

Framework feature lists are difficult to compare. Implement one bounded workflow and require every candidate to pass the same recovery and control tests.

Keep fixture evidence labelled

The agent controls lab uses authored framework outcomes.

from agent_cases import framework_case

result = framework_case()
assert len(result["acceptance_tests"]) == 5
assert result["eligible"] == ["framework_a"]
assert result["selected"] == "framework_a"
assert result["frameworks_executed"] is False
print(result["scope"])

Framework A passes all five fixture tests; B passes four. No framework runs, and setup times are invented.

Define acceptance before implementation

Test typed state, checkpoint/resume, approval pause, idempotent write and total timeout. Add observability, cancellation, deployment fit, licence and team support. Keep model, tools and task cases identical.

Measure code/configuration complexity, testability, cold start, persistence dependencies and failure recovery—not only demo speed. Use official current documentation because APIs change.

Select architecture, not brand

A plain state machine may satisfy the task with less dependency risk. If a framework wins, record version and the feature that justifies it. Avoid coupling domain logic to orchestration internals; keep tool and state contracts portable.

The Generative & Agentic AI course teaches one implemented framework while comparing alternatives through a design record.

Exercise

Implement the same five-step workflow in two current runtimes and plain Python. Execute 20 failure tests, measure setup/recovery evidence and allow the plain workflow to win.

Continue learning

This article is part of the Agent workflows and state sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the prerequisite or neighbouring task in Build a support triage workflow with an escalation path.
• Continue with Write an agent runbook for partial completion.

References: LangGraph documentation and Microsoft AutoGen documentation.