From Centralized AI to Embedded AI Agent Teams
AI agent teams are small groups of engineers and domain experts who build software agents that sit within day‑to‑day business workflows, automating repetitive tasks, connecting multiple systems, and supporting decision‑making in areas like HR, finance, and legal by working alongside staff rather than replacing entire departments. This shift matters because it puts AI where the work is done, not in a distant technical silo. Instead of one centralized group trying to automate everything from afar, companies are starting to embed AI talent directly into non‑technical teams so they can observe messy, real processes and turn them into working business process AI tools that genuinely save time. In practice, this is changing how enterprises think about workflow automation and who owns it.
Uber’s recent move makes this trend hard to ignore. Praveen Neppalli Naga, the company’s tech chief, embedded 30 of its “most AI-proficient engineers” into finance, legal, and human resources teams over a two‑week period to build tailored AI agents for specific business functions. These embedded teams, branded as agentic pods, have already run 16 separate deployments in the last two months. This is not a lab experiment; it is a deliberate reconfiguration of enterprise workflow automation, and it signals that AI’s real impact may arrive first in back‑office processes rather than shiny consumer features.
What Uber’s Agentic Pods Reveal About Enterprise Workflow Automation
Uber’s agentic pods show that meaningful enterprise workflow automation starts with proximity, not architecture diagrams. Over two weeks, these AI engineers sat inside HR, finance, and legal, watched how work truly happened, and then created AI agents to help handle those tasks. Many of the workflows they targeted, such as financial pacing reports, were not elegant flowcharts but sprawling routines that touched multiple systems and demanded large amounts of manual effort. In Naga’s words, “You can't automate them effectively by looking at process diagrams or documentation. You have to understand how the work actually gets done”. That quote should make every CIO uneasy about any AI program built primarily from PowerPoint and process maps.
The payoff has been stark. Those same financial pacing reports now take about 10 minutes to produce, down from two days, once AI agents handle the cross‑system data gathering and formatting. Allocating capital across 150 cities, a task that previously consumed 15 hours of staff time, now takes roughly 30 minutes with these agents running the heavy lifting. These are not marginal tweaks; they are step‑changes in how multi‑departmental business processes operate. The agents are not only stripping away repetitive data work but also shaping faster, more consistent decisions, because the information is assembled the same way every time. In effect, enterprise workflow automation is moving from static scripts to living AI colleagues.
Agentic Pods and the New Cross-Functional Team Philosophy
Uber’s agentic pods echo a broader rethinking of team structure driven by AI. As AI accelerates coding, Andrew Ng has argued that the bottleneck no longer sits in engineering but in every function surrounding it—from product management to marketing, legal, and design. His response has been to organize very small, high‑context, highly empowered generalist pods that blend technical depth with cross‑functional responsibility, allowing engineers to use AI to draft legal terms, write marketing copy, and move work forward without waiting in line for other departments. The underlying logic is simple: when AI makes individual contributors much faster, having large, functionally siloed teams stops making sense. A generalist pod can “run like crazy and build and ship code” while also creating something reviewable for specialists later.
Uber is applying a similar philosophy to business process AI, but pushing it into non‑technical domains. Instead of asking HR or finance to submit requirements to a central AI group and wait, Uber drops AI engineers directly into those teams, gives them context and autonomy, and expects them to redesign work from the ground up. Naga has said that Uber is now forming a dedicated team to scale this agentic pod model and go deeper, with the goal of using AI to “fundamentally change how the business operates”. In practical terms, that means HR staff who co‑create agents for onboarding, finance teams who own their capital allocation agents, and legal teams that partner with engineers on contract review flows. The team boundary stops being “technical vs business” and becomes “humans and AI agents aligned on a shared workflow.”
The Cost Question and Why Business Process AI Will Spread Anyway
The obvious pushback is cost. Uber’s leaders have publicly admitted that the company has ramped up spending on AI and even maxed out its Claude Code budget for the year, while not seeing a matching surge in “useful” consumer features. That tension matters: enterprise AI projects that impress internal teams but do little for users invite hard conversations about priorities. Yet the numbers from these agentic pods suggest the real value may lie in invisible workflow gains, not headline products. Cutting a two‑day reporting process down to 10 minutes and shrinking a 15‑hour capital allocation task to 30 minutes is the kind of improvement that compounds across quarters. Once these agents exist, they keep saving time every time the process runs. Over thousands of runs, the hidden return may outweigh the upfront AI budget, even if it never shows up as a flashy feature in an app.
There is also a talent signal buried in this story. While many tech firms are cutting jobs, they are still hiring forward‑deployed engineers whose role is to sit with customers and build AI agents tuned to their workflows. That looks very much like the agentic pod model: AI‑proficient engineers embedded in non‑technical settings, turning tacit know‑how into automated systems. Given Ng’s view that AI‑accelerated generalist pods will create new categories of work rather than eliminate them, enterprises should expect more positions that blend engineering with process design, and more teams that treat AI agents as standard teammates. The bigger risk now is not over‑investing in AI, but under‑investing in the organizational changes needed to make that spend pay off.
Conclusion: AI-Augmented Workflows Will Redraw the Enterprise Org Chart
The lesson from Uber’s agentic pods is clear: if you want AI to change your business, you must change how your teams are organized. Embedding AI engineers directly into HR, finance, and legal groups, having them observe real work, and asking them to build agents that make that work faster and easier has produced time savings that traditional automation efforts rarely deliver. At the same time, Andrew Ng’s pod philosophy shows that similar logic applies to product teams: small, high‑context AI‑powered generalists can move faster precisely because they cross functional boundaries. Put together, these examples signal a shift from AI as a separate capability to AI agent teams as a core part of enterprise workflow design.
In the next phase of AI adoption, the enterprises that win will not be the ones with the largest language models or the flashiest demos. They will be the ones that treat business process AI as a team sport; that embed technical talent into the heart of HR, finance, and legal; and that accept the discomfort of redesigning work in partnership with AI agents rather than bolting them onto old processes. Agentic pods are an early expression of that future. They show that the org chart itself is now a design problem for AI—and that the companies willing to redraw it will unlock the biggest gains in enterprise workflow automation.
