Uber CTO Praveen Neppalli Naga is back with a different AI lesson, after earlier “tokenmaxxing panic” fuel in the industry. This time, he is describing a concrete operating method: Uber is sending its best AI engineers into departments like finance, legal, marketing, customer support, HR, and procurement to study real workflows before building AI software.

Naga calls these two-week teams “Agentic Pods.” The process is simple but ruthless: engineers spend the first days shadowing employees, learning every step of a workflow, then quickly build and test software to improve it. And according to Naga, the results are not just incremental. A financial planning process dropped from 15 hours to 30 minutes, financial reports fell from two days to 10 minutes, and marketing quality checks shrank from two weeks to less than an hour.

That is the pivot. The headline number is eye-catching, but the real point is why Naga says most value does not come from shaving seconds off a single task. He argues the biggest gains instead come from redesigning entire workflows around AI, eliminating unnecessary approvals, replacing old software, and helping people make decisions faster. In other words, the AI is not the product. The product is the changed workflow.

If you have spent time watching large tech teams build AI tools, you know how hard it is to get from “cool demo” to “process actually changes.” Workflows are sticky. People have incentives to keep old handoffs. Systems have institutional memory. Legal and compliance requirements can turn “automation” into “extra review,” not less. By embedding AI engineers inside finance, legal, and procurement for two weeks at a time, Uber is compressing the feedback loop between model capability and operational reality. You can think of it as borrowing a page from the popular “forward-deployed engineer” concept. But Uber is doing the deployment internally, not outward to customers.

The twist is exactly that, and it matters for how leaders should interpret the move. Multiple tech executives compared the approach to Silicon Valley’s fast-growing “forward-deployed engineer” role. The source then highlights a new label applied to this internal version: Peter Wilczynski, chief product officer at Vantortech, joked about the “Rearward Deployed Engineer,” describing engineers embedded deep inside corporate operations to redesign workflows around AI rather than simply speeding up individual tasks.

From a decision-maker perspective, this kind of embedded pod model is also a governance play. When AI touches approvals, reporting, and decision-making, the risk is not just performance. It is control: Who reviews what? Which systems are considered authoritative? How do you document changes? How do you ensure the AI workflow does not quietly create audit problems later? Even though the source does not go into regulatory specifics, it does frame the departments included. Finance, legal, HR, and procurement are the places where internal controls, audit trails, and policy constraints typically live. That makes the “shadow the workflow first” approach more than a productivity trick. It is a way to align AI experimentation with the actual structure of compliance and operational responsibility.

There is also a capital and budgeting implication hiding in plain sight, especially given the reference to the earlier “tokenmaxxing panic.” The newsletter context notes that the earlier panic was sparked after an executive said Uber had already blown through its 2026 budget for Anthropic’s Claude Code. Whether you call that spend careful or reckless, the tension is familiar: boards and finance leaders want measurable outcomes from AI spend, not just raw usage. Agentic Pods look designed to answer that question with process metrics you can put on a slide. When a planning cycle collapses from 15 hours to 30 minutes, it is easier to argue ROI than when the claim is “the model is smarter.”

Second-order, this deployment method can reshape how teams are staffed. Instead of routing all AI talent into a central product org and handing departments an output tool, Uber is reversing the center of gravity. The AI experts go to the workflow owners. That changes training, incentives, and even how you measure success. It also changes what “AI strategy” looks like operationally, because the scope shifts from building software features to redesigning business processes around them. If Uber keeps finding time savings like the finance and marketing examples cited by Naga, other companies will be forced to ask whether their own AI programs are too detached from day-to-day work.

For peers in similar roles, the strategic stake is straightforward: if you are investing in AI, you cannot treat workflows as optional. You either redesign the workflow so humans can make decisions faster with the help of AI, or you accept that AI will remain an expensive add-on. Uber’s CTO is effectively saying the winning move is to take two weeks, go inside the process, and build only what improves the whole system. In an era where AI budgets and execution pressure collide, that is the lesson that reads like an advantage, not an experiment.