I was on the opening keynote panel at the LION Independent News Sustainability Summit today in San Diego — a session called “Humans all the way down: The risks, ethics and opportunities of AI in local news.”
I’m not a developer — Almost all of my part came out of one project: automating the record of local government meetings. The Moab Sun News now houses records of 179 meetings and 8,722 individual votes, going back to January 2023 — from the Grand County Commission, Moab City Council, and both planning commissions, including two bodies nobody else covers. No subscriptions, mostly automated and transcription costs about 38 cents per hour of audio.
One of the main points I want to get across is that for a lot of tools people might want to build, the result is mostly ordinary software — a shell script, a transcription service, 278 regular expressions, a database table, a hook that unpacks the votes and renders them everywhere. By line count, roughly 58% of the meeting flow is plain deterministic code.
What AI did was let me write that code. Using AI is not the same thing as prompting a chatbot every time and publishing what comes back, and the difference goes further than people expect: you can use a model to build a hard-coded system that does what you want, and then not use a model in an ongoing way at all. It was scaffolding. Not only does that mean that you don’t rely on your claude or chat gpt subscription forever, but deterministic code also doesn’t drift or produce unexpected results like using an LLM model every time might.
The other two arguments I made. First, transcription is a commodity — you can buy it anywhere, and on its own it’s of marginal use. What makes the output worth publishing is everything local expertise can hand the model before it writes a word: for us, a roster of about 370 name and organization corrections I built by hand, so the transcript is fixed before the model ever sees it (there are six recorded misspellings of just one commissioner’s name in that file); rich details from the agenda and the full staff packet, so that item titles, addresses, applicants and dollar figures come from the official document rather than from audio; and three years of prior meetings, so it knows whether an item is new or the fourth time a rezone has come back. The transcription is raw material — and like raw meat it’s not ready. The local context is the hot grill that makes it delicious.
I like having more info after a talk for people who are interested so I wrote up how and why to write a context document for your organization. After that I’d go to technical sources rather than to the AI-and-journalism conversation; the people I actually read and watch are on my resources page, which is short on purpose.
The notes I spoke from, with the research I cited and links to the meeting hubs, are at maggie-mcguire.com/lion. If you’re trying something like this in your own community and a piece of it isn’t working, I’d like to hear about it — maggie@moabsunnews.com.