Built This Week/AI by Industry

AI by Industry

AI in Government

We haven't hosted anyone from government, so this is the outside view: builders using public data, contractors mining agency records, and a defense supplier that won't ship AI code without proof.

We haven't had a government official on Built This Week, so this page is the outside view. AI shows up around government in three ways on the show. Builders turn open city data into tools: Jordan mapped every Los Angeles parking ticket. Contractors mine public agency records: Tristan Wilson's Edgevanta reads Department of Transportation bid history. And defense suppliers verify AI-modernized code: Ryan Aytay's Code Metal does this for the Air Force and defense contractors.

How companies we talked to use AI in government

Code Metal: verified code and war gaming for defense

Ryan said Code Metal is "trusted by the Air Force." It built a modeling and simulation platform with the military called War Matrix. War Matrix brings years of siloed data together and runs scenario simulations for war gaming. Code Metal also modernizes legacy code for defense companies that make software-defined radios, surveillance and field communications gear. One example is translating a million lines of C++, which isn't memory safe, to Rust, with a guarantee that it runs on the target device. (Episode 48, from 11:16)

So we have this belief, right, where, you know, generating code actually, the price is going down. It doesn't really cost anything, but if you get it wrong, it's pretty substantial.

— Ryan Aytay, Episode 48

Edgevanta: AI on public bid data

Civil contractors bid on public road and highway work. Tristan said Edgevanta goes to Department of Transportation websites, gathers historical pricing, and runs agents on top. The agents answer where the market is for a job, how much work competitors have, and what's likely on upcoming bids. Jordan's closing hope was that governments adopt AI on their side too, to write better contracts and diagrams: "AI talking to AI." (Episode 16, at 12:07 and 27:03)

What we built

Jordan's Tix LAX maps where parking tickets are issued across Los Angeles. He saw a viral live map of San Francisco tickets, which the city reportedly kept blocking, and found LA publishes its data through Excel, queries and APIs. He plotted it on Mapbox and coded almost all of it with Cursor and OpenAI Codex. The map has per-officer coverage, a leaderboard (officer 56 led with 329 tickets) and filters by car color, violation and agency. (Episode 15, from 01:38)

I think this is just like the beginning of, you know, giving, you know, residents and citizens access to their city's data to to be able to look at it in different ways.

— Jordan Metzner, Episode 15

For Edgevanta, Jordan also built two games, Edgevanta Tycoon and a highway paving mini-game. Both were for fun.

What's still hard

  • Data isn't live. LA's ticket data ran about two days behind, so the real-time officer tracking from the San Francisco map wasn't possible. The data also had future-dated tickets, likely mistyped years.
  • Cities push back. Jordan said San Francisco took its parking system down several times to block the live map, citing officer safety.
  • "Mostly correct" isn't enough. Ryan said AI-generated code "never will be" safe enough by itself for mission-critical systems at the edge, which is why defense work needs formal verification.
  • Who oversees frontier models. In news segments, Jordan said a model as dangerous as Anthropic described "maybe" is a national security issue (Episode 38). Guest Zach Rattner said AI is reaching "adolescence" and he'd welcome clearer guidance. Jordan was less sure regulation is the answer (Episode 49).

Jordan also questioned whether parking enforcement pays for itself. He said he'd read that LA's parking division was cash-flow negative, and the per-officer totals in his data looked small for a city that size.

Episodes on AI in government

FAQ

How is AI used in government?

On our show it showed up around government rather than inside it: Jordan used AI coding tools to map LA's open parking-ticket data, Edgevanta's agents analyze state DOT bid history for contractors, and Code Metal verifies modernized code for defense customers.

Can you build apps on open government data with AI?

Yes. Jordan built Tix LAX, a map of Los Angeles parking tickets, from city data available through Excel, queries and APIs, coding almost all of it with Cursor and OpenAI Codex. The data ran about two days behind, so live tracking wasn't possible.

How does AI help with public infrastructure bids?

Edgevanta pulls historical pricing from Department of Transportation websites and runs agents on top so civil contractors can see where the market is for a job and what competitors are doing.

Is AI-generated code safe for defense systems?

Code Metal's Ryan Aytay says not on its own. AI code is mostly correct, so Code Metal verifies translations with formal methods before they run on mission-critical hardware.