10 / AI / Governance
AI safe enough for the whole team
An 11-skill content governance system that encodes compliance and brand rules, so every teammate's AI output is safe by default.
- Rules
- In people's headschanged toEncoded in 11 skills
- Safe output
- Depended on the userchanged toDefault for everyone
- Review load
- Every draft by handchanged toHundreds of hours back
- Regulatory risk
- Carried by the reviewerchanged toDesigned out
Challenge
Once a team has AI tools, everyone becomes a publisher. In a regulated category that is a problem, because the rules about what you may claim are detailed, market-specific and easy to breach by accident. A model asked to write persuasive copy will write persuasive copy, and persuasive copy is exactly where non-compliant claims come from.
Training people to remember the rules does not scale, and reviewing everything by hand puts one person in front of every piece of output.
What I did
- Encoded the rules instead of teaching them. Eleven skills, each covering a defined job: listing copy, ad copy, customer replies, blog content, comparison content, outreach and so on.
- Wrote the compliance and brand constraints into the skills themselves, so the boundary applies at generation time rather than at review time.
- Made the safe path the default path. A teammate with no compliance training gets output that already respects the rules, because the rules are in the tool rather than in the prompt they had to remember.
- Kept the brand voice in the same place, so consistency and compliance are maintained together rather than fighting each other.
Result
- Eleven skills covering the content the team actually produces.
- Compliant, on-brand output by default for everyone, including people who never read the regulation.
- An estimated several hundred hours a year of review and rewriting capacity returned to the team.
- Regulatory risk designed out of the workflow rather than caught at the end of it.