How It Works

1. Structure process knowledge

Capture source systems, examples, constraints, and business intent. Break the process into bounded stages. For each stage define inputs, outputs, acceptance criteria, and who is responsible for approval. The result is an execution brief that an engineer can inspect.

2. Execute using AI agents

Provide the brief to an appropriate agent within the approved environment. Use AI to assist with explanation, transformation, troubleshooting, or test creation. Record the instructions, assumptions, and generated artifacts. Stop for human judgment when requirements are ambiguous.

3. Validate results

Check outputs against independently defined acceptance criteria. Use baseline comparisons, automated checks, and human review. Validation is a gate: failures return to the relevant execution stage with an explanation, rather than passing silently to the next step.

4. Deliver measurable outcomes

Preserve reviewed code, evidence, decisions, and handover instructions. Compare observed results against the agreed baseline using workload-specific measures. Document limitations and assign ownership for maintenance. No performance result should be reported without supporting evidence.

The relationship between the stages

Inputs and context ? bounded AI execution ? validation and human review ? approved outputs. Review can send work back to context gathering or execution. Human oversight defines the acceptance criteria and authorizes progression throughout the loop.

Use the workflow catalog to find documented methods. The SAS migration solution illustrates the approach for a specific technical problem.