Use this to design a workflow that actually holds up
Before You Start
This worksheet is for thinking through a workflow, not generating one automatically.
Do not use it to produce a copy-paste output by vibes.
Use it to force clarity before you build.
If you are designing an agent, orchestrator, reusable prompt system, or internal AI workflow, this is meant to help you think through the architecture deliberately.
0. Anchor the Artifact
What final thing should this workflow produce?
Real example artifact:
If no example exists, describe the target output:
Why is this considered good?
What must not be lost in reproduction?
Deployment mode:
one-off
repeatable personal workflow
reusable operator module
Stop if: the output is still vague.
0.5 Pattern Locks
What makes the example actually work?
Structural locks
What process moves must be preserved?
Surface locks
What quality / tone / rhythm / texture must be preserved?
Anti-patterns
What should future outputs avoid?
Stop if: you still can’t explain why the example works.
1. Use Case Bind
Exact use case:
Final deliverable:
Human role:
AI role:
Success condition:
Unacceptable failure condition:
Stop if: someone else wouldn’t understand the job.
2. Constraints
Hard constraints
Soft preferences
Must-preserve conditions
Forbidden failure modes
Time / effort tolerance
Reversibility requirement
Stop if: you don’t have at least one hard constraint and one unacceptable failure.
3. Runtime Reality
Guaranteed inputs
Optional inputs
Forbidden assumptions
Allowed tools / sources
Volatile inputs
Continuity assumptions
Insufficiency triggers
Stop if: you’re still designing for context the workflow won’t actually have.
4. Failure Profile
Likely shortcut behavior
Hidden-state / fake-success risks
Stage-mixing risks
Evidence / assumption laundering risks
Highest-cost wrong answer
Stop if: you haven’t named how the workflow is most likely to fail.
5. Architecture Shape
Candidate shapes
Why each fits or doesn’t
Selected shape
Reason for selection
Stop if: you chose the shape because it “sounds smart.”
6. Stage Construction
For each stage, define:
Stage name:
Objective:
Allowed evidence:
Prohibited moves:
Required artifact:
Advancement gate:
Collapse condition:
Recovery move:
Repeat for each stage.
Stop if: any stage has no artifact, no gate, or no collapse behavior.
7. Reliability Attack
Try to break the workflow.
Could the AI jump to the final answer early?
Could a stage be faked?
Could evidence boundaries be violated?
Could the final output look right without process compliance?
Are any transitions narrative rather than conditional?
Are any pattern locks merely decorative?
Will repeated use cause formulaic output?
Weaknesses found
Fixes needed
Stop if: you haven’t actually tried to break it.
8. Final Package Check
Is this workflow heavier than the job requires?
Does every stage materially reduce failure risk?
Is any stage present only for elegance or ceremony?
Is the quality gain worth the overhead?
Is it appropriate for the deployment mode?
Final decision
finalize
simplify
roll back and revise
The One Question That Matters
When building the workflow, do not ask:
What are the steps?
Ask:
What artifact must exist before the next move is allowed?