Intent Engineering

Turn user friction into AI-ready shipping specs.
From the first complaint to a verified shipped feature — structured, repeatable, agentic.

Friction

→

Spec

→

Execute

→

Verify

→

Close Loop

01 · Capture Friction

Collect raw signals from users. Support tickets, interviews, analytics drop-offs, NPS comments, session recordings. Cast wide — filter later.

Spec Title

Product / Area

Status

Friction Signals — What are users complaining about or struggling with?

Severity Frequency Weight = S × F (max 25)

02 · Structure Into Spec

Write the five-part intent spec. Every field makes the spec executable by an AI agent without follow-up questions.

Objective — The user problem and why it matters

User Goal — The job-to-be-done

Outcomes — Measurable success criteria

Edge Cases — Boundary conditions and failure modes

Verification — How to confirm it works

03 · Execute With Agents

Hand the spec to an AI coding agent. A well-written spec produces a working implementation on the first pass — or surfaces exactly where the spec was ambiguous.

Generated Spec — Ready to Hand Off

Agent / Tool

Execution Notes — What the agent produced, questions it raised, blockers

04 · Verify & Ship

Use the verification criteria from your spec to confirm the implementation works. Check each item before calling it shipped.

Add Verification Check

Ship Notes

05 · Close the Loop

After shipping, check whether the outcomes you specified actually materialized. This is the signal that feeds your next specs.

Friction Signals

Evidence Weight

Verify Items

Items Checked

Status

Draft

Outcome Review — Did the outcomes you defined actually happen?

Learnings for Next Spec — What would you write differently?