The Agentic AI Workflow Canvas: 12 sections
The Canvas is the main deliverable of AI 104. It puts everything from Modules 1 to 5 on one page, so a workflow can be designed, discussed with others and handed to the next stage of implementation.
Core question
How do we design a whole agentic workflow clearly, communicate it, and hand it to the next stage?
| # | Section | What it records |
|---|---|---|
| 1 | Goal | What the workflow should accomplish |
| 2 | Trigger / Input | What starts it: a request, an uploaded file, a schedule, a new record or an external event |
| 3 | Task Decomposition | Major tasks and subtasks |
| 4 | Execution Order | Sequential, Parallel, Routing or Loop |
| 5 | Decision Points | What must be decided, with what information, and whether by rule, model, agent or human |
| 6 | Memory, Knowledge & Skills | For each step, the memory, knowledge and skill it needs (from AI 103) |
| 7 | Tool / Action Steps | Where an external action or data source is needed, such as search, database, email, calendar, API or an enterprise system |
| 8 | Execution Output | What each step produces |
| 9 | Verification | How results are checked: rule, reference source, second model pass, comparison or human review |
| 10 | Human-in-the-Loop | Review, Approval, Escalation and Override |
| 11 | Failure / Fallback | What happens when information is missing, a result is invalid, a tool fails or confidence is low |
| 12 | Final Output | The final deliverable |
Info
You do not need to design tool schemas or write any code. Section 7 only marks where a Tool Step is needed. That marked list becomes the input to AI 201.