The Last Mile Just Got Shorter.

In trucking, the "last mile" is the most expensive part of every delivery. The long haul is efficient. Highways are predictable. But that final stretch -- navigating neighborhoods, finding doorsteps, dealing with dogs and stairs and humans -- that is where the cost lives.

Yesterday, the last mile got shorter.

The Task

DoorDash launched a standalone app called "Tasks." It pays delivery drivers to record themselves performing everyday activities. Folding clothes. Washing dishes. Making beds. Pruning plants. Recording conversations in Spanish with friends and family.

The footage trains AI and robotics models to "understand the physical world." Not just for DoorDash. The data feeds models developed by partners in retail, insurance, hospitality, and technology. DoorDash, a company built on human labor navigating the last mile, is becoming an AI training data broker -- using the same workforce the trained models will eventually replace.

The program is available to 8 million U.S. gig workers. It excludes California, New York City, Seattle, and Colorado. Those happen to be the states with the strictest gig worker protections and data privacy laws. Draw your own conclusions.

We Have Seen This Before

This pattern is not new. It is just moving down the economic ladder.

It started with knowledge workers. Law firms paid attorneys to annotate legal documents that trained AI models to do legal research. The attorneys did the work. The models learned. The attorneys got laid off. Companies like Mercor now hire those same displaced lawyers, scientists, and PhDs as gig workers to produce training data -- at a fraction of their former salaries.

IBM paused hiring for 7,800 positions and asked remaining employees to document their processes for AI handoff. Google restructured its ad sales division after employees spent months training an AI system called "Gemini for Sales" on their own client relationship workflows. A 2026 Gartner survey found 64% of organizations used existing employees to create AI training data. Only 22% were transparent about how that data would affect future staffing decisions.

Now it is gig workers. People already operating without benefits, job security, or negotiating power. Filming themselves washing dishes in their own homes for a few dollars per task, training robots that will eventually navigate the last mile without them.

As one analyst put it: "It is a bit like assembling the robot that might eventually take your shift."

The Sugar in the System

There is a pattern in how industries optimize for human behavior, and it does not always optimize for human welfare.

Food companies learned decades ago that adding sugar to everything -- bread, pasta sauce, yogurt, cereal -- made products more desirable. Not more nutritious. More desirable. The products sold better. The companies grew. The long-term costs showed up in healthcare systems, not quarterly earnings reports. The optimization target was consumption, not health.

The AI labor pattern runs on the same logic. Companies need training data. The cheapest, highest-quality source of training data is the people who currently do the work. So you pay them to document how they do the work. The task feels like extra income. The worker benefits in the short term. The long-term cost shows up somewhere else -- in displaced livelihoods, in communities built around work that no longer exists, in a generation that trained the systems that made their skills redundant.

The optimization target is model capability, not workforce stability.

In Los Angeles, former delivery drivers are already being hired as "robot wranglers" -- swapping batteries, wiping lidar sensors, freeing stuck robots from potholes, completing deliveries when the robots die mid-route. The pay is $21 to $23 an hour. That is $45,760 a year in a city where MIT calculates the living wage for a single adult at $63,402.

The last mile got shorter. The gap between what workers earn and what they need did not.

What This Means for Coordination

At OTP, we think about how AI agents coordinate inside organizations. The DoorDash story is what happens when that coordination question gets scaled to the entire economy.

When one company deploys AI agents, the coordination challenge is internal: which agent owns what, how do they hand off, what are the authority boundaries. When every company deploys AI agents simultaneously, the coordination challenge becomes structural: who works, who trains, who benefits, who decides.

The organizations that will navigate this well are the ones that make their coordination intelligence explicit. Not just the technical rules for how agents work together, but the human-AI boundary conditions -- where automation stops and human judgment begins. We track these as knowledge claims in every OOS published on this platform, because the organizations that document those boundaries are the ones that will hold them.

The ones that do not document them will discover the boundaries after they have already been crossed.

The Other Side of the Transition

Here is what is easy to miss when you are inside the disruption: transitions end.

The industrial revolution displaced millions of agricultural workers. It was brutal for the generation that lived through it. Their grandchildren lived in a world with indoor plumbing, electricity, antibiotics, and a life expectancy that doubled. The internet wiped out travel agents, video rental stores, and classified ad revenue. It also connected 5 billion people to the sum of human knowledge and created entirely new categories of work that did not exist before.

AI will displace work. That is happening now and it will accelerate. The generation living through this transition will bear real costs -- lost jobs, stagnant wages, communities that need to reinvent themselves. That pain is not hypothetical. It deserves honest acknowledgment, not dismissal.

But the world on the other side of this transition is one where the drudge work -- the repetitive, dangerous, soul-crushing labor that humans do because someone has to -- gets handled by systems that do not get tired, do not get hurt, and do not have families to feed. A world where autonomous delivery means lower costs for everyone. Where AI-assisted medicine catches diseases earlier. Where the knowledge of how organizations coordinate their AI teams is shared openly, compared rigorously, and improved collectively -- which is what we are building here.

The last mile just got shorter. For this generation, that is uncomfortable. For the next generation, it might be the reason the things they need cost less, arrive faster, and free up human time for work that actually requires being human.

The transition is the hard part. We are in the transition. The organizations that get through it well will be the ones that were honest about what AI changes, deliberate about what humans keep, and structured about how the two work together.

That is the work.