It’s been three busy weeks, with a bit of time off in the middle, so there’s been a gap in my notes. This is a summary and reflection on the whole stretch.
Major achievements
- Reached a clear decision on using plug-ins. And how we’ll distribute Claude skills. Both decisions give teams a simpler route to consistent content guidance.
- Completed two skills. One for taxonomy and one for our component patterns so that teams can draft consistent copy with AI.
- Got our content roadmap live, giving teams and leadership visibility of what’s coming.
- Built an auto-updating accessibility requests list. It pulls in three sources of customer feedback and gives us one prioritised, evidence-based place to check.
Wins
- Captured our navigation patterns, which only lived in our heads. Now, anyone can apply them.
- Updated the quarterly content impact and effectiveness report.
- Documented the plug-in status and strategy to align the cross-functional group.
- Moved beyond traditional content audits by testing a reverse-engineered, AI-led approach.
- Started auditing a sign-up journey. Already caught errors, like a typo in which “your” was “you” in a key first-touch flow.
- Partnered with our content designer on a progress update for our owned areas.
- Tested a skill created to build presentation decks. Prepped feedback for its creator.
- Cross-checked the component skill against another skill to avoid overlap.
- Completed Anthropic’s AI Fluency course and internal training.
Opportunities
- Turn the AI-led audit approach into a repeatable playbook others can use to audit their own content.
- Decide where guidance outside the main content skill should live.
- Stress-test other skills and cross-check them so there’s no overlap.
- Finalise the accessibility user-testing research plan and get it reviewed.
- Extend the impact report to the wider team.
- Share the feedback on the deck skill with its creator.
Learnings
- Content knowledge that lives in people’s heads becomes much more useful once it’s written down as skills.
- Skills need to be cross-checked early to avoid overlap.
- We didn’t need a plug-in. Simpler distribution won once the group compared the options.
- Traditional audits aren’t the only route. An AI-led approach can surface issues fast and might become a repeatable method.
- Having Claude review as I build helps shape skills so they’re structured for agents.
That’s the last three weeks. Most of it came back to the same thing: getting what we know out of our heads and into something other people can use. Back to regular notes next week.
Leave a Reply