1
Describe the feature idea
Any format — rough note, Slack message, customer quote, stakeholder ask
Try an example:
Claims history
Onboarding checklist
Data export
Step 1 of 4 — runs one AI call
1
Intake Triage — structured problem statement
Review and edit before passing to Step 2
2
JTBD Framing
The structured problem above is passed as context — generating functional, emotional, and social job layers
2
JTBD Framing — jobs-to-be-done layers
Review before passing to Assumption Mapping
3
Assumption Mapping
JTBD output → key assumptions ranked by uncertainty × impact
3
Assumption Map — ranked by risk
Review before generating the prioritization brief
4
Prioritization Brief
All prior outputs combined → RICE scoring rationale + recommended next action
4
Prioritization Brief
RICE scoring rationale and recommended next action
How this pipeline works
Design decisions at each step
Step 01 — Intake Triage
Why structure first, not frame first
Raw ideas arrive in any shape — quotes, Slack messages, vague asks. Before any framing can happen, the AI needs a consistent schema to work from. Step 1 extracts: the core problem, who is affected, and what they cannot do today. This gives every downstream step the same clean inputs regardless of how messy the original idea was.
Step 02 — JTBD Framing
Why JTBD runs on Step 1's output, not the raw idea
If you ask an AI to generate a Jobs-to-be-Done statement from a raw stakeholder quote, it will reflect the stakeholder's framing — not the underlying job. Step 2 receives the structured problem statement from Step 1 as its input, forcing the JTBD to be grounded in the actual constraint, not the solution request. The three-layer output (functional, emotional, social) surfaces dimensions that never appear in a feature request.
Step 03 — Assumption Mapping
Why assumptions come before scoring
RICE scores are only as good as the assumptions underneath them. Most prioritization fails because teams score features as if their assumptions are already validated. This step explicitly surfaces what must be true for the feature to matter — ranked by the combination of uncertainty and impact. High-risk assumptions become spike candidates; low-risk ones can proceed to backlog.
Step 04 — Prioritization Brief
Why the final call uses all three prior outputs
Most AI tools use a single prompt and call it a pipeline. This step demonstrates genuine chaining: the prioritization brief is constructed from the structured problem (Step 1), the JTBD layers (Step 2), and the assumption risk map (Step 3). The RICE rationale is contextualized by the jobs and the risks — not scored in a vacuum. The recommended next action is constrained to four valid options: spike, prototype, backlog, or reject.
Chain Design
The human checkpoint principle
Each step presents its output before the next one runs. This is deliberate. An agentic pipeline that runs without human review accumulates errors silently — a bad framing in Step 2 produces a misleading assumption map in Step 3 and a wrong recommendation in Step 4. The step-by-step design is the safety architecture, not a UX decision.
Progress: