AI Workflow Definition Review
Definition. A structured review of a workflow's design before build, covering the happy path, every branch, every failure mode and every compensation. Design review is far cheaper than fixing a workflow in production, and it is where most defects can be prevented.
How it works. Walk through the definition with the business owner and an engineer. For each step, ask what happens if it fails, if it is slow, and if it runs twice. Verify that every branch has both outcomes and every step has a compensation where needed. Check that the failure paths are testable. Record the review and the decisions.
Example. A design review of a payment workflow asks what happens if the refund step times out. The answer is that the workflow retries, and the review reveals the refund is not idempotent, so a retry would double-refund. The step is redesigned with an idempotency key before build. The defect would otherwise have been discovered in production, as it had been in a previous workflow.
Use case. Design review is required for any workflow with financial or irreversible steps, and it is the cheapest quality practice available. It is also how you transfer knowledge of failure modes between teams.
Warnings. A review that only walks the happy path misses the defects that matter. Never skip the question of what happens on retry. And never let the review be a formality signed off without the business owner present.
Limitations. Reviews take time and require participants who understand both the business and the failure modes. They also cannot anticipate every failure, so testing remains necessary.