Uber's Agentic Pods and the Open Loop
Uber just published the most concrete enterprise AI story I have seen. Its CTO described “Agentic Pods”: embed a handful of the company’s most AI-fluent engineers directly into a business team, finance, legal, HR, for two weeks, have them watch the actual work, and build AI agents alongside the people doing it. Not a central AI team working from a requirements document. Not off-the-shelf tools handed to each department. Builders sitting next to the business, building with them rather than for them. Sixteen pods in two months, and the numbers are the kind that stop a meeting: a financial pacing report from two days to ten minutes, capital allocation across 150 cities from fifteen hours to thirty minutes, marketing QA from two weeks to under an hour.
Uber got the hard part right, and it is worth naming exactly what that is in precise terms.
What Uber actually proved
Strip the story down and Uber ran an experiment on a specific interaction: a business owner who holds the problem and the budget, working directly with a small, senior, AI-fluent building pod. In RACE Programming that interaction has names. The business owner is the Team Principal: the person on the client side who owns product direction and decides what is worth building. The building pod is the Pit Wall: the client-facing pair, a Forward Deployed Engineer and an AI Product, who sit with the business, turn what it needs into a specification, and stand up a working prototype fast.
That is the front of the loop, and Uber validated it better than any framework deck could. The instinct to embed senior AI-augmented people next to the business, and to build with the people who own the problem, beats the two default plays: a central AI group translating tickets, or a tool licence dropped on every desk. Most large companies have not figured this out. Uber did.
The loop is still open
But a pod that stops at the Pit Wall produces agents. Point automations that sit beside the company’s core software and speed up a task around it. That is real value, and it is where most enterprise AI stops today.
What is missing is the back half of the loop: the Pit Crew. In RACE Programming the Pit Crew is the execution unit, a Pit Crew Quality Engineer and Pit Crew Software Engineers, that takes the specification the Pit Wall produced and ships it as production software, every Stint, through the Definition-of-Done gates, with the Silicon Software Engineer, the AI agent, doing the mechanical build inside a team that owns delivery end to end. Close the loop, Team Principal to Pit Wall to Pit Crew, and the requirement the pod discovered does not dead-end in a script. It flows into an Executable User Story and out as a change to the actual system.
Why closing it matters
An open loop builds agents and agentic applications. A closed loop builds those and the classic enterprise systems a business runs on: the systems of record, the core applications, the software that has to keep working and keep evolving. Those systems are never done. They need continuous development, and, crucially, they need someone close enough to the business to know what the next change should be. That is precisely what an embedded pod learns while it watches the work. The same closeness that lets a pod automate a task tells a full loop what the enterprise system should become next.
That is the difference between reengineering a department’s process and building the company’s software. One produces automations around the system. The other produces the system, and keeps producing it as the business changes.
Uber proved the pod, which is the front of the loop, the Team Principal working with the Pit Wall. The step the story leaves open is to close the loop by adding the Pit Crew. Do that, and you have a single end-to-end modern engineering approach that delivers not only agents and agentic applications, but the enterprise systems underneath them, evolving as fast as the business does. RACE Programming is that closed loop.
Written by Pavel Khodalev, author of RACE Programming and CTO of First Line Software. Follow new essays via RSS.