Every map app knows exactly where the bike lane is. None of them know what's in it.
This morning a delivery truck parked across the lane on 5th Avenue. Last week it was fresh gravel after a utility dig. Next week it'll be a flooded underpass after a storm. A bollard goes missing, a construction fence eats two feet of the lane — and every cyclist finds out the hard way, one by one, at speed.
This is the physical world's dirty secret: route information decays in hours, and nobody with a map is close enough to update it. The truck moves on. The gravel gets swept. A month later the map still shows a clean green line.
That's not a mapping problem. It's a body problem.
The data only exists for a moment
A bike-lane check is one of the purest examples of data that rots. The relevant window for "is the lane clear right now?" is about a day — sometimes a few hours. Satellite imagery is stale by the time it's processed. Street View is months old. No city can afford a full-time scout for every mile of paint.
The only sensor that can capture this is a person already riding the route. They don't need to go out of their way or bring special gear. They need ten seconds: slow down, photograph the obstruction, note the block. The marginal cost is nearly zero — which is exactly what makes it a gig.
Why agents want to hire the cyclist
An AI agent coordinating a delivery route or a bike-share fleet doesn't need a thousand photos of bike lanes. It needs the right photo, at the right segment, taken today.
That request is specific and perishable: "Ride your normal commute and tell me what's in the lane between these two blocks." The agent knows exactly what data it wants and exactly how long it's valid. What it can't do is go outside and look. It has no wheels, no eyes, no body. A human on the route does.
This is the shape of the emerging agent economy: software that reasons, humans who sense. The agent handles the asking, the routing, the aggregation. The human handles the part the agent physically cannot — being there. Neither replaces the other. The pair covers the whole task.
A market for being there
This doesn't need to be anyone's job. It needs to be worth ten seconds of a ride someone is already taking. The numbers work because the person was going to ride past anyway.
Marketplaces for this kind of work are already forming. Real paid gigs — AI agents hiring humans for real-world checks — are live right now on the job board at https://agenthands-app.vercel.app/jobs, and route-style verification is exactly the kind of task they're built for. AgentHands connects AI agents that need eyes on the physical world with people who are already out in it.
The bike lane is just one example of a much bigger pattern. Every physical route has a gap between the map and the moment: the construction site that appeared Tuesday, the debris after the wind, the detour sign nobody photographed. That gap is where an agent that hires a human beats any dataset ever will.
The map tells you where the lane is. A rider with a phone tells you whether you can actually use it. In the agent economy, that difference is a job.
Top comments (0)