Be honest about the starting point: no policy degree then, no network, a Substack nobody read. The only asset was the ability to pull public data and explain it. Room of about 65, including people who work in government: say the pipeline works from inside an agency too.

This is the whole talk on one slide. The next four slides walk each stage with what actually happened and when. The arrow at the end loops back: everything I learn inside becomes the next post.

Pick two of the four to tell as stories; the unfunded mandates post is the one that proves the pipeline, because a former DOT commissioner read it and forwarded it to the agency's leadership within four days. Method, if asked: AI pulls and parses the data, I check its work, it never writes my conclusions.

The writing and the data are one project: a post surfaces a dataset, the dataset becomes a tracker or an explorer, and the tracker produces the findings the next post is built on. Say it out loud: AI collapsed the build time. One person can now ship what used to take a team a year. The hard parts are the ones government still has: procurement, security review, maintenance. If there is time, pull up one tracker page live. The tool is the business card: agency staff share a tool with colleagues in a way they never share an op-ed.

This is the G train post from September 2026 on every surface. The order matters: the table and charts exist before the writing starts, and they outlive the post as tools other people use. The thread and the video are cut from the same charts, so one week of data work feeds four channels.

The point for the class: one of these was an application. The posts did the rest of the applying. Name Sarah Sachs and Polly Trottenberg out loud if it feels right in the room; the slide keeps roles.

Each of these is a question a reader could have asked at a community board meeting, answered with public data and every chart open. The audience is modest in size. What these posts did was reach the right people.

COGE and the mandates post are the same move: take a government document everyone cites and nobody reads, turn it into a dataset, and show agency by agency what it changes. The board guides fill the same gap for community boards.

Tie to Daniel's thesis explicitly. The board seat is unpaid and the work is unglamorous, and it is the single best source of questions I have. Stories, pick two: the hour-long debate on whether a park bench faces the playground or the street; the vote on a zoning map change to recognize that a stretch of the Allen Street median is already a park (not to create one), so Parks could renovate the building in it; being called out for writing the CB reports as the board went to a vote on firing an assistant district manager. Then the monthly reports, the AI review of a full meeting, the scorecard from 2,409 recordings.

The point of this slide: none of these were planned. Each came out of a conversation the writing started, with a person who then became a collaborator. The lesson that follows is "talk to people."

This is the closing slide before thanks: end on the ask that they pick one of these and do it before the next class. Say which two of these you'd have skipped and regretted. [TAL: still your call which lessons stay; this is the draft list.]