Serve Robotics Opens Its 2,000-Robot Sidewalk Fleet to All Developers

Serve Robotics is opening its fleet of 2,000 autonomous sidewalk delivery robots to outside software developers, small businesses, and even autonomous AI agents through a new AV Delivery Portal (its developer interface for dispatching robots via API). The move, announced by the company, turns a fleet that until now has served primarily major food delivery platforms into what its CEO describes as an open marketplace — one that anyone in the 150 neighborhoods where Serve operates can summon directly.

The stakes are economic, not technical. Delivery apps such as DoorDash and Uber Eats take a 15 percent to 30 percent cut on every order, largely because paying human drivers is expensive. By letting merchants connect straight to its robots through software, Serve allows bakeries, florists, and pharmacies to add robot delivery to their own websites and checkout registers without paying those middleman fees.

Why is Serve opening up now?

The developer platform arrives on the heels of Serve's announcement that it does not plan to renew its partnership with Uber Eats, echoing similar moves across the transportation industry to build independent platforms rather than depend on dominant aggregators.

Serve, which was spun out of the app-based delivery company Postmates in 2021, has made Uber Eats deliveries in Los Angeles, Miami, Dallas, and dozens of other cities for several years, using robots that look like autonomous shopping carts. Opening the platform reduces that dependency. It also solves a utilization problem unique to food.

In an interview with The Verge, Serve CEO Ali Kashani explained the demand pattern that shapes his company's economics:

"The reality of food delivery platforms is that usually they dominate their markets. Food is generally very picky. So you have lunch, you have dinner. But then there are times in between. So again, having that diversification helps."

Lunch and dinner rushes leave robots idle. Off-peak hours are dead weight on the balance sheet. Spreading demand across the day — and across use cases far beyond restaurants — is how Serve expects to push operating costs down.

What does the developer portal actually offer?

Third-party developers, independent businesses, and autonomous AI agents can use Serve's software tools to dispatch robots directly, without routing orders through a major delivery app. The mechanics are split cleanly between the two sides:

• Developers supply the pickup and dropoff coordinates.
• Serve's back end orchestrates obstacle avoidance, routing, and remote teleoperation (human supervisors monitoring robots remotely).

There is no queue-jumping for enterprise customers, Kashani said. API requests and enterprise orders are treated identically:

"There's no preferential treatment at the end of the day. Until the robot is on a delivery, it doesn't have a commitment to anybody."

Pricing comes in two models: pay-per-delivery or hourly reservation. For small merchants, Kashani says both will undercut human-powered delivery services:

"This would be a fraction of that cost."

Who is this for?

The target audience extends well beyond restaurants. Serve's pitch covers:

• Local shops — bakeries, florists, pharmacies — adding delivery to their own websites and checkout registers.
• Intra-office shuttles, where businesses program robots to carry items between offices instead of hiring couriers or sending junior staff across town.
• Repair shops ferrying parts.
• Peer-to-peer exchanges like Facebook Marketplace.

Anyone living in one of the 150 neighborhoods where Serve currently operates can employ a robot for a delivery. That geographic number matters: the platform's reach is real but bounded, and expansion depends on where Serve chooses to deploy hardware next.

Why does the CEO frame this as an economics crusade?

For Kashani, opening fleet access is the next phase in a long-standing campaign against last-mile delivery economics. He frames the core inefficiency in one question:

"Why does getting something delivered from China cost $1, but from Chinatown cost $10?"

The answer, in his telling, sits in labor costs and intermediary fees stacked on short-distance trips. Kashani connects lower per-delivery costs directly to fleet utilization and scale, arguing that expanding demand beyond peak restaurant hours into off-peak windows and new use cases drives overall operating and hardware costs down for everyone.

The economics cut both ways. Lower utilization means each idle robot-hour is spread across fewer paying deliveries, raising effective cost per trip. Higher utilization — filled by developers, small merchants, and AI agents during off-peak hours — spreads the same hardware and operations spend across more revenue. That is the arithmetic behind the open platform.

Are there limits on what the robots will carry?

Yes, and they are manual. Sidewalk autonomy remains governed by the laws of physics and by Serve's own gatekeeping. Access to the robots will be strictly gated by manual approval from Serve's employees, a measure intended to prevent hazardous deployments or illicit cargo. Liability on stolen or broken goods will be subject to how individual partnerships are priced and insured, Kashani noted.

The safety profile, however, is modest by design. The robots have a top speed of around 6 mph, which Serve says means little to no risk to pedestrians. The company also argues the robots are a safer alternative to cars and help reduce traffic congestion, since they operate mainly on sidewalks rather than roads.

Do not expect the form factor to grow. Kashani remains adamant that Serve will not become a street-based delivery service anytime soon. The 50-pound sidewalk robots will not morph into full-size vehicles. If software developers want to automate the sidewalk, Serve will gladly sell them the wheels.

What does this mean for the delivery market?

The immediate effect is competitive pressure on the app layer. A local shop in one of Serve's 150 neighborhoods can now theoretically offer delivery at "a fraction of that cost" of human courier services, without surrendering 15 to 30 percent of each order to a platform. If developers build on the portal in volume, the 2,000-robot fleet becomes shared infrastructure rather than a dedicated Uber Eats subcontractor — a position Serve has now formally exited by declining to renew the partnership.

The longer question is utilization. Kashani's argument that off-peak demand from offices, repair shops, and peer-to-peer senders can drive down per-delivery costs for everyone rests on developers actually showing up. The portal, the pricing models, and the open API are the bet; the lunch-and-dinner gap is the opportunity; and the 6-mph shopping cart is the vehicle.