On July 7, 2026, Uber CTO Praveen Neppalli posted something you rarely see on X: the exact, step-by-step operating method behind how a company that size is pushing agentic AI beyond engineering. They call it Agentic Pods, and in two months they’ve already run 16 of them across 16 different business functions, with results you can verify in the same thread.

This isn’t a PR post with vague numbers. It’s the CTO of a public company laying out, with a concrete 10-day sequence, exactly how they structure the work of pairing AI with a business function that isn’t engineering. And what he describes matches, point for point, the two-month discovery we run in AI Maestro.

What an Agentic Pod actually is

By Neppalli’s own account, Uber already had massive AI adoption inside engineering: 99% of its engineers use AI tools, more than 70% of pull requests come from local or cloud agents, and its engineers have built more than 2,500 agent skills across the software development lifecycle. The question they asked next was different. How do you bring that beyond engineering? Finance, legal, operations, marketing, customer support, HR, procurement.

The answer was forming roughly 30 Agentic Pods, pairing one of their most AI-proficient engineers with a domain expert from the business function in question. Every pod got exactly 10 days, on a fixed sequence that doesn’t vary. Days 1 and 2, the engineer shadows the expert, observes every step, documents the workflow, asks questions. Day 3, they prioritize opportunities by scale, repetition, business impact, and data availability. Days 4 and 5, they build a working agent alongside the person doing the job, not for them. Days 6 through 9, they validate with several other people doing the same work, to confirm it generalizes and actually makes the job better. Day 10, they ship it.

The results, and the line that matters most

In two months, Uber ran 16 Agentic Pods across 16 different functions. Some concrete numbers straight from the thread: capital allocation across 150 cities went from 15 hours to 30 minutes. Financial pacing reports went from 2 days to 10 minutes. Marketing web content quality assurance went from 2 weeks to 50 minutes. Support workflow creation went from 9,000 manual workflows to self-service.

But the productivity gain wasn’t what surprised Neppalli most. What he highlights is how quickly engineers embedded in an unfamiliar domain found opportunities that had been hiding in plain sight. And one line sums up the whole pattern: “the workflow becomes the unit of automation, not the individual task.” The biggest wins didn’t come from automating a single task, they came from redesigning the entire workflow around the agent, eliminating handoffs, removing unnecessary approvals, replacing legacy tooling.

Why most companies can’t just copy this

Here’s the part almost nobody mentions when they share this thread. Uber’s method works because Uber has something most mid-market companies in Central America don’t: roughly 30 engineers with deep AI fluency available to lend out for 10 full days to a department that isn’t their own. That’s an internal talent bench built over months of prior investment in AI adoption inside engineering.

If your company doesn’t have that internal bench, Uber’s 10-day sequence doesn’t automatically translate into results. You need someone with that same profile, capable of sitting with the business expert, documenting the real workflow (not the process diagram nobody actually follows), and building the agent alongside the person doing the work. That’s exactly what we solve as an outside partner in AI Maestro discovery: the same pattern of pairing, shadowing, prioritizing, and building that Neppalli describes, just run by someone from outside when the company doesn’t have the internal engineering bench Uber does. We’ve written before about the 95% AI utilization gap: the tool is already available, what’s missing is exactly this kind of structured process to find where to point it.

The biggest lesson Neppalli left wasn’t a productivity number. It was this: the best AI opportunities are rarely visible from outside the function where they live. You find them by sitting next to the person doing the work, understanding every point of friction, and building with them, not for them. Uber has the engineering bench to do that 16 times in two months. Most companies need someone else to do it once, done right.