Multi-Tool AI Workflow Brief
Use this before combining AI tools, search tools, databases, documents, and automation platforms.
Workflow Summary
- Workflow name:
- Owner:
- Business outcome:
- Users:
- Source of truth:
- Final output destination:
Steps
| Step | Tool | Input | Output | Human check | Failure fallback |
|---|---|---|---|---|---|
| 1 | |||||
| 2 | |||||
| 3 |
Source Status
Label each source before it enters the workflow.
| Source ID | Type | Version / period | Authority | Sensitive data? | Approved tool/account |
|---|---|---|---|---|---|
| Public / internal / customer / restricted | Primary / secondary / unverified | Yes / no |
Handoff Rules
- What context must be carried from step to step:
- What should be summarized rather than copied:
- What must never leave the source system:
- What is the single canonical record after the workflow completes:
Quality Gate
- Required checks before publishing or acting:
- Minimum evidence required:
- Reviewer or approver:
- Stop condition:
Run Log
- Date:
- Operator:
- Tools used:
- Links to source records:
- Result:
- Follow-up:
Claim Handoff
| Claim ID | Checked observation | Source IDs and calculation | Limit preserved downstream |
|---|---|---|---|
| C1 (invented example) | Week 2 count rose 25% | S1=40, S2=50; (50-40)/40 | Counts do not establish productivity or cause; incomplete Week 3 excluded |
- Recipient can access the required evidence without expanded permission:
- Claim rejected during review and reason:
- Source changes that require the output to be rechecked:
Record observations, inferences and decisions separately. An AI-generated source packet still needs verification before it is labeled checked.