If you run a state or local government agency, you're being told to do something about AI. The reflex is a plan, whether a task force, a roadmap, or a year-long strategy. 17A Partner Zack Wortman and I wrote this paper to argue for a much smaller first move. Pick one piece of real work and start.
The ask is intentionally modest. Pick one accountable owner, get five to ten paid licenses for Codex or Claude Code, and give the team 90 days. At the end you will have a working tool, a team with trust and expertise, and a quantifiable metric for the value they created.
What's in it
- Why you're ready now: Government workers already use AI about as often as private-sector workers (Gallup, late 2025). The gap is organizational.
- The foundation: Four things you put in place once, none of which needs meaningful new budget or new authority
- First pilots: A checklist for picking use cases, a week-by-week walk through the first ninety days, and three worked examples
- What goes wrong: How pilots stall from the inside, and how to get past security, legal, procurement, and workforce objections
Who it's for
Directors, division heads, and program leads in state and local government, and the IT and security teams they need on board. It assumes old systems, a thin technical bench, and no AI policy, because that describes most agencies.
Take it with you
The full paper is public (link above). It closes with five things a director could do this week, starting with naming an owner and asking managers one question: what recurring work eats the most time for the least judgment? Do that much and you've started.
Background
This grew out of 17A's AI enablement work with public-sector teams. The advice is specific to government, down to small-purchase thresholds and FedRAMP, because that is where generic AI advice tends to fall apart.