September’s AI Community of Practice session closed out the MAP series we started back in May: the last step, Polish, plus an open discussion where public health peers compared notes on governance, access, and where the real friction is right now. If you couldn’t make it, here’s what came out of it.
Watch the recorded portion of the session on YouTube. A transcript is available on the video page.
It’s not prompt and polish. It’s just polish.
We opened with a fast recap of Map and Add Context for everyone joining us for the first time, then got into the real subject of the day: the last step of the framework. We used to call this step “prompt and polish.” After sitting with it for a few more months, that name doesn’t hold up. It’s about the polishing.
“A great prompt gets you a great first draft. It doesn’t get you a finished one.”
The “brilliant intern” metaphor from Add Context came back up, because it still holds: AI shows up with a PhD in everything, and it doesn’t know your values, your jurisdiction, or your community. A strong prompt gets you most of the way there. Getting it over the finish line is still your job.
We also talked about what Silicon Valley has started calling “meat proxies”: people who take whatever an AI model hands them and pass it along without actually reading it. That’s the failure mode polish exists to prevent. Before anything goes out, ask three questions: is this true? Does it sound like us? Is it actually ready to go out?
From the discussion: governance is the hard part, not the technology
The open discussion that followed surfaced a few recurring themes from public health departments working through their own AI rollouts:
- Context beats cleverness. One participant shared turning a rambling, in-the-moment voice memo into a clean SOP, because giving the model everything she was thinking, unfiltered, beat a tidy prompt with less context.
- Grant compliance is where AI earns its keep quietly. One department described uploading grant documentation and having AI catch a staffing-commitment discrepancy, a contract that had promised more FTEs than were actually assigned, before it became a compliance problem.
- Access policy is unsettled everywhere. Several departments are still working out who should have access to which tools, and what happens when the sanctioned tool falls short: people find workarounds, which is its own governance risk.
- Records retention rules vary a lot by state, and departments are building AI governance frameworks without much shared precedent to draw on.
- Sustainability is a real open question for grant-funded roles. If your AI workflows live in your own personal accounts, what happens to that work when the position, or the funding, goes away?
That last point ties back to something we believe pretty strongly: use AI for the things that don’t require your heart or your head, and build that into infrastructure the department owns, not habits that live in one person’s account. If your department is working through the access and governance questions above, AI Policy Development is where we help teams build that framework directly, in a facilitated session where your team makes the decisions.
One format change starting in October: we’re moving to cameras-on for the discussion portion. This group stays small and off the record on purpose, and seeing each other is part of what makes people comfortable saying the real thing.
The AI Community of Practice runs through the year, with a break over the summer. It’s a free, open conversation for people working in or adjacent to governmental public health who are thinking seriously about AI. Learn more and join the community →
