AI doing small jobs well.

Why I keep this list

I choose to be an optimist about AI. I believe it can raise the quality of what we make and give more people work that is creative, fulfilling and useful. The other path is the dystopian one: cheaper and cheaper goods, more layoffs, and a shrinking workforce.

Optimism is a direction, not a forecast. This list keeps mine honest. Each entry shows what the AI actually does, where a person makes the call, and what the best evidence says, including when the evidence says "not yet."

Case study

Cameras on school buses

Verdict: promising, not proven

When a school bus stops and its stop arm swings out, a camera on the side starts recording. Software flags any car that drives past and reads its plate. A vendor employee screens the flag, a police officer approves or rejects it, and a civil ticket goes to the car's owner. The AI does the first sort. People decide who gets a ticket. (NHTSA, 2024, B)

The problem

In one day in spring 2026, about 101,000 bus drivers in 39 states and DC counted 56,024 cars illegally passing their stopped buses. (NASDPTS, 2026, B)

Deaths are rarer than that number suggests, and most aren't caused by passing cars. From 2015 to 2024, 77 school-age pedestrians died in school-transportation crashes. A school bus hit 44 of them, and other vehicles hit 33. (NHTSA, 2026, A) In the 2024-25 school year, all six deaths reported in school bus loading zones were caused by the bus itself. (Kansas State Department of Education national survey, 2025, B)

Does it work?

Cameras catch far more drivers than police or bus drivers can. Whether they change behavior is still open. Two federally funded studies, in five districts across four states, could not show a clear drop in passing. (NHTSA, 2021, B; NHTSA, 2024, B) In the 2024 study, the camera operator shared only combined totals, so the researchers couldn't check its numbers. No company publishes how often its system flags the wrong car.

What went wrong in Dallas

A camera company's CEO admitted paying local officials more than $3.5 million to win Dallas County Schools' stop-arm camera contracts. Dallas County Schools was shut down in 2017 with about $103 million in debt. (US Attorney, Northern District of Texas, 2019, A) The lesson isn't about cameras. It's about money: when someone's income depends on more tickets and more contracts, watch the incentives.

Here in Colorado

Colorado allowed these cameras in May 2025. The fine is capped at $300, the school board has to approve, and the vendor's pay can't be based on fines or on ticket quotas. (Colorado HB25-1230, A) As of November 2026 I found no Colorado program issuing tickets yet.

What would make it a clear good use

• Share raw data with independent evaluators, so someone without a stake can measure the effect.
• Publish how often the system flags the wrong car, and how often police reject a flag.
• Keep footage only when there's a violation, and only for enforcement.
• Look at the bus too. Most children killed at bus stops are hit by the bus itself, so a system that sees children in a bus's blind spots may save more lives than any ticket.

Going deeper: paying for the good

If the cameras work, fewer drivers pass, fines fall, and the camera company earns less. I think that's what success looks like, as long as nobody depends on the fines. I wrote a short essay on why, with an interactive model you can adjust and a list of who to read.

Read the essay and try the model