At New England Drupal Camp 2026, I gave a talk called Trusted AI: Guardrails, Governance, and Proof That Your Team Still Owns the Work. It opened with four questions.
We ask them before we start any AI work at Kanopi. And that covers more than just code. This also applies to internal workflows, content, client work, and even the new tool someone found on Friday.
We call them PREP:
- Purpose: What outcome does this support?
- Risk: What could go wrong, and who would feel it?
- Evidence: How will we know it works?
- Person: Who is the named person responsible six months from now?
Try this prep before you adopt a tool, a workflow, or a feature.
Purpose: What outcome does this support?
P is for Purpose. What outcome does this support? “Use AI more” isn’t a purpose. A purpose sounds like: more test coverage on a legacy module. First drafts of alt text for an editor to review. Discovery interviews summarized so that strategists spend their time on findings rather than transcripts. If you can’t name the outcome, you can’t tell whether AI helped, and you can’t tell when to stop.
Risk: What could go wrong, and who would feel it?
R is for Risk. What could go wrong, and who would feel it? Three kinds come up again and again.
- Data: client or user data going into a tool that keeps it or trains on it.
- Quality: confident, wrong output that ships because it looked finished.
- People: the site visitor, the editor, or the client who trusts the result.
An incorrect internal summary and an incorrect answer on a health client’s public site are not the same risk, and they shouldn’t be handled the same way.
Evidence: How will we know it works?
E is for Evidence. How will we know it works? Decide what good output looks like before you start, and write it down. While the work happens, check against it: tests, evals, and a human review.
Afterward, keep a record of what was checked and by whom. Evidence lets you say yes to a tool with confidence, instead of crossing your fingers. It’s also what you’ll show a client who asks how you use AI.
At Kanopi, that record is a Delivery Record. It names the person who is responsible for the work and the checks they ran.
Person: Who is responsible six months from now?
The final P is for Person. Who is the named person responsible six months from now? That person reviews, approves, and answers for the result. And they have the authority to change how the tool is used, or to stop using it.
AI can do the work. A person owns the result.
The slides and the leave-behind (PDF) are both online. Let me know if you implement PREP into your workflows.