Waymo's paying DoorDash drivers $11 to close robotaxi doors in Atlanta. A passenger takes a ride, gets out, doesn't fully close the door, and the vehicle just sits there—stuck. The sensors won't let it move with an open door, so Waymo pings nearby Dashers, offering $6.25 plus a $5 bonus to walk over and shut it. Waymo asks DoorDash drivers to close doors

This is what full autonomy looks like in 2026: a six-figure self-driving car that can navigate city traffic, read traffic lights, and handle complex intersections, but can't close its own door.

The companies frame it as win-win. Waymo keeps its fleet moving, Dashers get a quick micro-task layered on top of normal deliveries—another example of how gig platforms create flexible earning opportunities in unexpected ways. It's presented as efficient labor allocation—dynamic routing of humans to wherever the robots get stuck. Waymo-DoorDash partnership

But this isn't Waymo's first attempt to patch the door problem with humans. In Los Angeles, they've been using the Honk app—basically Uber for tow trucks—to dispatch operators who get paid $20-24 just to close a stuck door, and $60-80 if the car needs an actual tow. Waymo pays tow truck operators The DoorDash pilot is the cost-optimized version. Instead of a tow truck blocking traffic, grab the nearest gig worker on foot and treat it like a two-minute side quest.

What's really happening here is Waymo's running into an unglamorous hardware constraint. Their current fleet uses Jaguar I-Pace EVs with traditional manually-closed doors. The software won't move if sensors detect a door as open or obstructed—which means a dangling seatbelt, a lazy close, or a slammed-but-not-quite-latched door is enough to strand the car. Tow truck operators rescue robotaxis

Waymo says future platforms will have automated door systems—sliding subway-style doors instead of hinges. Hyundai's even patented double-sliding doors specifically for autonomous vehicles. In other words, the long-term fix is "change the car," but the short-term fix is "invite more humans into the loop." Waymo's future automated doors

Here's what bugs me about this: we're not replacing human labor with autonomy. We're reassigning it. Instead of driving the car, you're paid to unstick it. Instead of being the driver, you're an on-demand field technician for the autonomy stack. The job didn't go away—it just became smaller, weirder, and less predictable.

I'm pro-Waymo. They've proven autonomous vehicles work at scale—100,000+ rides per week across multiple cities. That's not a pilot. That's infrastructure. But this door-closing pilot shows the gap between "autonomous driving" and "autonomous operation." Waymo can handle the main task—driving—better than most human drivers. But all the edge cases around that task? Still human labor.

What we're seeing is a template for the future of work: automation handles the "main" job, and a marketplace of micro-tasks emerges around all the failure modes. Gig platforms become the glue layer, routing humans to wherever robots get stuck. For workers, that means ultra-granular tasks—jobs that last two minutes and pay a few dollars, mixed into a day of conventional gigs.

For cities, it means roads full of highly capable self-driving vehicles that still depend on an invisible layer of human labor to stay functional. You won't see that labor because it's distributed across thousands of gig workers getting pinged with $11 tasks between DoorDash deliveries. But it's there, patching the gaps that autonomy can't yet close.

The irony is hard to miss: in the race to build fully autonomous systems, we're also building new ways to sell human time in ever-smaller slices. Waymo's stuck doors are the most literal version of that tension—autonomy that still can't quite close the loop without someone walking over and giving it a shove.

Here's what I think happens: Waymo rolls out sliding doors on the next vehicle generation, the door-closing micro-tasks disappear, and everyone forgets this was ever a problem. But the template remains. Every new autonomous system will have its own set of boring, annoying edge cases that require human intervention. And every time, companies will solve it the same way: spin up a gig marketplace, pay people $10-20 per task, and treat human labor as the elastic band that keeps automation from breaking.

That's not a failure of autonomy. It's just what autonomy looks like when you zoom in close enough to see the humans still holding it together.