Uber Embeds AI Engineers in Business Units to Build Custom Agents

Uber's embedded engineering experiment

Uber is testing a new model for deploying artificial intelligence across its organization by embedding small teams of AI engineers directly within business departments. The company calls these units "agentic pods."

According to CTO Praveen Neppalli Naga, Uber placed 30 of its most AI-capable engineers with teams in finance, legal, human resources, and other functions. Over two-week sprints, these engineers shadowed employees, observed their daily workflows, and built custom AI agents to handle repetitive tasks. The company has completed 16 such pods over the past two months, Naga said in a post on X.

The approach targets work that involves accessing multiple systems and significant manual effort—tasks that resist automation through standard process documentation alone. Financial pacing reports that previously required two days of work now take 10 minutes with the new agents. Capital allocation across Uber's 150 operating cities dropped from 15 hours to 30 minutes.

Why it matters

Uber's pod model addresses a persistent challenge in enterprise AI adoption: the gap between technical capability and operational reality. By requiring engineers to observe work firsthand rather than rely on process diagrams, the company is building agents that reflect how tasks actually get done, not how they're supposed to be done on paper. This embedded approach may offer a template for other organizations struggling to translate AI investment into measurable productivity gains.

The forward-deployed engineer role

The agentic pod structure mirrors a job category that has remained in demand even as tech companies cut other positions. Forward-deployed engineers—technical staff who work directly at customer sites or within business units—represent a rare growth area during recent industry layoffs, Business Insider reported in May.

Naga emphasized that understanding real workflows requires direct observation. "You can't automate them effectively by looking at process diagrams or documentation," he wrote. "You have to understand how the work actually gets done."

Spending questions remain

The efficiency gains come against a backdrop of scrutiny over AI spending. In May, Uber COO Andrew Macdonald said on a podcast that justifying the company's AI expenditures had become more difficult. Naga told The Information that Uber exhausted its annual budget for Claude Code—an AI coding assistant—by spring. Yet that spending hasn't produced a corresponding increase in consumer-facing features that Macdonald described as "useful."

Despite these cost concerns, Uber plans to expand the pod model. Naga said the company is forming a dedicated team to scale the approach and "fundamentally change how the business operates" by redesigning work processes around AI capabilities.

These details were first reported by Business Insider.