Mold
Name the repeatable shape.
Trigger, source context, decision, action, owner and business measure.
Practical method
Turn expert work into leverage.
Start with how the firm actually works. Add AI where source-grounded preparation and measured follow-through can improve the outcome without blurring responsibility.
Pour
Fill it with verified context.
Meetings, email, documents, relationship history, permissions and approved examples become usable evidence.
Gem
Install human-controlled infrastructure.
The workflow prepares, proposes and learns; an accountable expert owns consequential action.
Allocate AI participation
- Automate
- Only bounded, verified and low-consequence work with clear exception handling.
- Prepare
- Collect and structure context for expert review.
- Recommend
- Present options, rationale and uncertainty; never approval.
- Human Decides
- Name the person or committee responsible for consequential judgment.
The control loop
Sense→Prepare→Expert approves→Act→Measure→Learn
Automation is earned by proof. If the source is weak, the action is consequential or the measure is unclear, the system pauses for a person.
Expert-Led AI Value Test
Measure the work, not the novelty.
- Define done. What exactly counts as a completed, usable outcome?
- Calculate full cost. Include model cost, operator time, review, rework and exception handling.
- Track dependability. How often is the output usable without a human fixing or finishing it?
- Keep tracking. Models and workflows change; value must be rechecked over time.