Uber burned through its entire 2026 AI budget in four months. The ride-hailing giant’s heavy use of tools like Anthropic’s Claude Code sent costs soaring. Engineers loved the productivity boost. Yet executives struggled to point to new customer features or clear returns. COO Andrew Macdonald voiced the frustration in May. ROI questions hung heavy.
Now the company has a different answer. Business Insider detailed how Uber CTO Praveen Neppalli Naga shifted tactics. Instead of flooding the organization with coding assistants, Naga handpicked about 30 of the firm’s most AI-proficient engineers. He embedded them directly into business teams. Finance. Legal. HR. Marketing. Procurement. Customer support. The goal wasn’t more code. It was observation followed by action.
These short-term squads, dubbed Agentic Pods, last two weeks each. Engineers shadow employees from day one. They watch every click, every system switch, every manual approval. No process diagrams. No static documentation. Real work only. “You can’t automate them effectively by looking at process diagrams or documentation,” Naga said, according to Business Insider.
The results came fast. Financial pacing reports that once took two days now finish in 10 minutes. Capital allocation across 150 cities shrank from 15 hours to 30 minutes. Marketing web quality assurance dropped from two weeks to under an hour. Support teams turned thousands of manual workflow creations into self-service options. And the numbers keep coming.
From Budget Panic to Workflow Redesign
Earlier this year the token-spending frenzy dominated headlines. Uber’s rapid adoption of AI coding tools consumed its annual budget by spring. Monthly costs per engineer ranged from hundreds to thousands of dollars. Ninety-five percent of engineers used the tools at least monthly. Roughly 70 percent of committed code came from AI assistance. Yet Macdonald noted no corresponding surge in shipped customer features.
That disconnect forced a rethink. Naga’s team moved beyond individual task acceleration. Pods target entire workflows. They eliminate redundant approvals. They replace legacy systems where possible. They surface data for faster decisions. The approach echoes Silicon Valley’s forward-deployed engineers who sit with customers to build tailored solutions. At Uber the talent deploys inward. Peter Wilczynski, chief product officer at Vantortech, called them “rearward deployed engineers” in a lighthearted jab that captured the internal focus.
Uber ran 16 such pods across 16 different business functions in a two-month span. Each pod pairs one AI-savvy engineer with domain experts. The first days focus on immersion. Questions fly. Friction points surface. By the end of week two the team ships working agents into production. No long pilots. No endless committees. Speed matters.
Executives at other firms took notice. The model addresses a common enterprise complaint. Central AI teams build platforms that business units ignore. Business units request features that miss daily pain. Pods collapse the distance. Engineers see the work. They feel the context. They ship solutions that stick.
Recent coverage reinforces the momentum. A July 15 analysis on Havlek.ca highlighted how embedding AI talent inside finance and operations teams produced commercially credible outcomes. Time savings translated into measurable capacity gains. One operations workflow that previously tied up multiple staff now runs with minimal oversight. Similar stories appear in discussions across tech forums and executive briefings from the past week.
Yet challenges remain. Scaling the pod model demands a steady supply of engineers who combine deep AI skills with business curiosity. Not every company has Uber’s bench. Cost control still looms. Even successful agents consume tokens. Uber itself plans a dedicated team to refine and expand the approach. “We’re now forming a dedicated team to scale this further and go deeper,” Naga stated. “They’ll deeply understand the work, redesign it from the ground up, and use AI to fundamentally change how the business operates.”
The shift signals broader movement in enterprise technology. Companies once chased chat tools and copilots for every desk. Many discovered limited impact on core operations. Workflow redesign demands proximity. It requires engineers willing to sit in finance meetings, watch legal reviews, observe HR onboarding. Tacit knowledge rarely appears in org charts or slide decks. Pods force its capture.
Uber’s experiment builds on its earlier AI investments. The company already reports thousands of agent skills built internally. Most engineers interact with AI daily. The pods represent the next layer. They turn experimental coding assistance into operational infrastructure. They move AI from the engineering org into the profit centers.
Analysts predict more firms will test similar structures. A July 9 piece from The State of AI described the pods as evidence that AI value now lives beyond engineering departments. Shadowing, mapping, building, validating, shipping. The cycle repeats. Each iteration refines both the agents and the understanding of how the company truly functions.
Critics point to measurement difficulties. Time saved does not always equal dollars earned. Some workflows resist full automation due to regulatory needs or judgment calls. Uber retains human oversight where required. The pods themselves include validation steps with multiple operators before deployment. Risk management stays central.
Still, the early metrics impress. Support workflow creation that once generated 9,000 manual items now offers self-service paths. Capital decisions accelerate. Reporting cycles compress. These changes free employees for higher-value work. They reduce error rates tied to repetitive tasks. They create capacity without headcount growth.
Naga’s pivot from budget alarm to pod evangelist offers a case study for technology leaders. Spend first. Measure later. Learn and adapt. The Agentic Pods concept strips away abstraction. It puts builders in the room where work happens. It treats the workflow itself as the product to improve.
Other organizations already study the model. Discussions on X in recent days highlight interest from CTOs across industries. Some experiment with their own versions using smaller teams or focused sprints. The core principle travels. Proximity beats speculation. Observation trumps assumption. Rapid delivery tests ideas faster than months of planning.
Uber shows no sign of slowing. The dedicated scaling team will likely spawn more pods, deeper integrations, and perhaps new platform capabilities drawn from repeated workflow patterns. What began as a response to runaway token costs evolved into a systematic method for discovering AI’s highest-leverage applications inside the business.
The ride-hailing leader that once disrupted transportation now experiments with disruption of its own internal machinery. Success here could influence how large companies approach AI for years. Failure would highlight limits of even the best engineers when dropped into complex domains. For now the numbers point up. The pods deliver. And the rest of the business watches closely.
Uber’s Agentic Pods: How Embedding Top AI Talent in Finance and HR Is Reshaping Enterprise Productivity first appeared on Web and IT News.