Uber's Agentic Pods: How 30 Engineers Cut 15-Hour Workflows to 30 Minutes

Two months later, Uber's CTO Praveen Neppalli Naga unveiled a radically different approach. The results were so striking — a financial planning process dropping from 15 hours to 30 minutes, reports from two days to 10 minutes — that it became one of the most closely watched enterprise AI case studies of the year . Here is how Uber's Agentic Pods work and why they represent a fundamental shift in how the company thinks about AI.

What Are Agentic Pods?

Agentic Pods are small, two-person teams — each pairing one AI-proficient engineer with one domain expert from a business function — given a fixed two-week sprint to observe a manual workflow and rebuild it with AI agents . Naga launched the program in mid-2026, deploying roughly 30 pods across 16 business functions including Finance, Legal, Marketing, HR, Customer Support, and Procurement over two months .

The Exact 10-Day Playbook

Naga published the precise workflow on LinkedIn on July 7, 2026 :

• Days 1-2: The engineer shadows the domain expert, observing every step of the real workflow.
• Day 3: The pod prioritizes the highest-impact task to automate.
• Days 4-9: The pair builds the AI agent collaboratively, validating it with the expert's peers.
• Day 10: The working agent is shipped into production.

The hard deadline is not incidental. It forces the team to ship something real rather than polish something indefinitely . As Naga described it, the goal was to "stop treating AI as an expensive experiment and start using it to redesign complex, manual workflows" .

Measured Productivity Gains

The results are widely reported across multiple outlets, attributed to Uber's internal data :

These represent cycle-time reductions of 30x to 400x for core business processes. Naga also reported that 9,000 manual support workflows had been converted to self-service .

From Tokenmaxxing to Targeted Automation

The earlier strategy — dubbed tokenmaxxing — was a spending-heavy approach where nearly 95% of Uber's ~5,000 engineers used tools like Claude Code and Cursor daily . But the company could not link that token consumption to any measurable business outcome. In May 2026, Macdonald told Business Insider it was becoming "harder to justify" AI spending because "that link [between token usage and productivity] is not there yet" . Higher token usage was not translating into proportional improvements in consumer-facing products .

The key differences between the two strategies:

• Tokenmaxxing era: Blanket deployment of AI coding tools to all engineers; spending tracked by token volume; no clear ROI link acknowledged by leadership .
• Agentic Pods era: Targeted, workflow-first approach; engineers embedded in non-engineering departments; success measured by concrete process time reductions .

Why the Pod Structure Worked

The Agentic Pod structure solved two problems the tokenmaxxing approach could not. First, it forced engineers to understand a problem end-to-end before building, rather than accumulating tokens without a defined output. Second, it tied AI spending directly to measurable business outcomes — hours saved, reports accelerated, workflows eliminated .

In just two months, 16 pods ran across 16 business functions . The model has since been expanded. As Naga noted, after the pods, 99% of Uber's engineers were using AI tools — but now with a framework to ensure that usage produced results the business could track .