TL;DR
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
A Pragmatic Engineer interview with Uber Distinguished Engineer Uday Kiran Medisetty details how the company’s developer infrastructure is being adapted for AI agents. Uber uses a model gateway, tools for switching coding harnesses, spending controls and AI-assisted review, while engineers still review every pull request.
The Pragmatic Engineer has published an interview with Uber Distinguished Engineer Uday Kiran Medisetty detailing how the company is connecting AI coding agents to its internal code, services and engineering knowledge. The account describes a model gateway, agent tools and spending controls, while emphasizing that human engineers still review every pull request.
Medisetty, described by The Pragmatic Engineer as one of Uber’s longest-tenured engineers and a leader in its agentic AI work, discussed a developer environment built around more than 20 internal tools. Some predate generative AI; newer systems let engineers work with coding agents while retaining shared settings and access to company infrastructure. Uber supports Claude Code, Codex and OpenCode through AIFX, a wrapper that carries settings, skills and plugins across those coding harnesses.
Uber’s Model Gateway, established in 2023, is designed to let teams switch among Anthropic, OpenAI and other vendor models, as well as models Uber runs itself. The interview says the gateway also supports security middleware that checks requests and responses, tracks latency, and provides spending records, audit logs and session traces. These measures give the company a common layer for managing AI use rather than tying each tool to one model provider.
The company also sets AI spending tiers for engineers. According to the interview, a manager can approve additional monthly spending once an engineer reaches a tier’s limit; the approach is intended to allow costly work when justified without encouraging wasteful token use. A plugin called CostCheck offers advice across sessions. The report says Uber has cut token costs by about 50% in recent months, but the supplied material does not give the measurement method or comparison period in greater detail.
How Uber Governs AI Coding
The systems described address practical questions for companies introducing AI agents into software development: which models can be used, how agents connect to internal systems, and who monitors cost and security. A shared gateway can make it easier to change providers, while common controls can give security teams and managers more visibility into usage.
Code review remains a key constraint. Medisetty told The Pragmatic Engineer that review has become a bottleneck, and Uber has built tooling to help engineers judge how much attention a change needs. The report says AI review runs before a human review, with supporting screenshots or videos added when relevant and risk scores assigned. Those steps may help reviewers prioritize work, but they do not replace human approval: all pull requests are still reviewed by engineers, according to the interview.
The account is relevant beyond Uber because it shows AI adoption as an infrastructure and workflow effort, not simply the rollout of a coding assistant. Provider flexibility, access control, spending governance and review processes all shape how agents can be used in a large engineering organization.
As an affiliate, we earn on qualifying purchases.
Infrastructure Built Before Agents
Uber’s internal tooling grew in response to the company’s scale and operating model. The interview recounts that Uber had two services when Medisetty joined in 2015, and thousands by 2019. The report says the number of services in 2026 is expected to be about the same as in 2019, despite the business being roughly ten times larger than it was then. Those figures are presented in the source as an account from the interview, not as independently audited measurements.
The company has previously built systems when commercial products did not meet its needs. Examples in the report include M3DB for metrics, Cadence for durable workflows and Morpheus for feature flags and experimentation. The source also describes cities as a core operating concept, reflected in areas such as service sharding and cloud-region placement. This established internal infrastructure helps explain why Uber’s AI effort focuses on connecting agents to existing systems and processes.
The 2023 decision to make AI-built systems model- and harness-agnostic preceded the current collection of agent tools. That design choice provides a way to change models or coding interfaces without rebuilding every integration. The interview presents this as a deliberate platform approach, rather than evidence that one particular model or agent has become the company-wide standard.
“Code review has become a big bottleneck inside the company.”
— Uday Kiran Medisetty, as quoted in The Pragmatic Engineer interview
As an affiliate, we earn on qualifying purchases.
Limits of the Reported Results
The interview provides a tour of Uber’s internal systems, but the supplied source material does not include independent performance evaluations or detailed usage data. The reported 50% reduction in token costs lacks a stated baseline, exact dates and methodology here, so it should be read as a company-reported figure rather than a comparable industry benchmark.
It is also unclear how frequently engineers receive additional spending approval, how the review risk scores are calculated, or whether AI-assisted review has reduced review times or defects. The source describes tools and practices, but does not provide measured results for those outcomes. It does not establish that every Uber engineering team uses each system in the same way.
As an affiliate, we earn on qualifying purchases.
How Agent Workflows Develop
The interview outlines current infrastructure rather than announcing a product launch or a dated rollout. Uber’s next steps are not specified in the supplied material. The systems described leave the company able to adjust model providers and coding harnesses, while continuing to track usage and route pull requests through engineer review.
Further evidence of the approach’s impact would include details on adoption across teams, changes in review time or code quality, and the methods behind reported token savings. Until Uber or the publication provides such data, the clearest confirmed takeaway is the design of the internal workflow: agents can assist with coding and review preparation, but engineers retain responsibility for reviewing pull requests.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did the report about Uber’s AI infrastructure describe?
The Pragmatic Engineer interviewed Uber Distinguished Engineer Uday Kiran Medisetty about internal developer tools, AI coding agents, model access, spending controls and code review.
Which coding tools does Uber support through AIFX?
The report lists Claude Code, Codex and OpenCode. AIFX carries settings, skills and plugins between those coding harnesses.
Does AI replace human code review at Uber?
No. The report says engineers still review all pull requests. AI review, risk scores and supporting materials are used before or alongside human review.
How does Uber control AI spending?
Engineers receive spending tiers, and managers can approve additional monthly spending after a limit is reached. The report also describes a CostCheck plugin that gives advice on usage costs.
What is not known about the reported token savings?
The source says Uber reduced token costs by about 50% in recent months, but the supplied material does not specify the baseline, exact period or calculation method.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
