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 describes how the company is connecting coding agents to its internal systems and engineering tools. The account covers Uber’s model gateway, spending controls and code-review workflow; the company says it has reduced token costs by about 50% in recent months.
Uber’s AI infrastructure combines a model gateway, coding-agent tools and internal developer systems, according to Distinguished Engineer Uday Kiran Medisetty in an interview published by The Pragmatic Engineer. The account offers a look at how the ride-hailing company is integrating AI into engineering work while retaining human review of pull requests and tracking model use and spending.
Uber built its Model Gateway in 2023 to let internal AI tools work across models and coding harnesses, rather than tie the company to a single provider. The gateway supports models from Anthropic and OpenAI, as well as open models Uber runs itself. The interview says security middleware called AI Guard checks requests and responses, while logging and session traces provide visibility into usage.
Engineers can use Claude Code, Codex and OpenCode through AIFX, a wrapper intended to carry settings, skills and plugins between harnesses. Uber also tracks AI spending by user and assigns engineers spending tiers. Managers can approve additional spending when an employee reaches a limit. A plugin called CostCheck offers suggestions for reducing costs across sessions.
The interview reports that Uber has cut token costs by about 50% in recent months. That figure is the company’s reported result; the material does not provide a detailed calculation, comparison period or methodology. The account also says AI code review runs before engineers review pull requests, with risk scores and, where relevant, screenshots and videos available to help reviewers judge changes.
How Uber Connects Agents to Engineering
The setup shows how AI coding tools at a large company depend on more than access to a model. A shared gateway can give teams a way to change providers, record usage and apply security checks across tools. That matters as engineering organizations weigh the benefits of coding agents against model costs, oversight and operational risk.
Uber’s account also describes a division of work rather than a handover to autonomous systems: agents can assist with code and review, but engineers still review every pull request. Risk scoring and supporting media are intended to help reviewers direct their attention, though the interview does not establish how much review time the tools save or how accurately the scores identify problems.
The company’s internal tooling reflects its scale and operating needs. Medisetty said Uber has more than 20 tools engineers use day to day. Systems built before today’s generative AI tools, including service ownership and observability infrastructure, form part of the environment agents now need to work within.
AI code review tools for developers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Service Growth to AI Tools
Medisetty joined Uber in 2015, when the company had two services, according to the interview. By 2019 it had thousands. The report says the service count in 2026 is expected to be about the same as in 2019, although the business is roughly 10 times larger than it was then. These are figures and comparisons presented in the interview, not an independently verified inventory.
Uber has previously built infrastructure when commercial products could not meet its requirements. The interview names M3DB, created to handle a reported 500 million metrics per second, and Cadence, a durable workflow engine. It also describes city-level requirements that shape deployment, observability and data systems: cities are associated with cloud regions and may fail over to another region.
The company’s AI tools were layered onto this broader developer environment. According to the report, Uber’s platform team decided in 2023 that AI development should be model- and harness-agnostic. The gateway and AIFX are described as products of that direction, allowing engineers to use different models and interfaces without rebuilding every shared control.
“Cities are a “basic primitive” that needs to be handled across systems.”
— Uday Kiran Medisetty, Uber Distinguished Engineer, in The Pragmatic Engineer interview
As an affiliate, we earn on qualifying purchases.
What the Interview Does Not Quantify
The interview does not provide independent measurements of whether coding agents have increased engineering productivity, reduced defects or shortened review cycles. It describes the tools and practices but does not give a company-wide before-and-after evaluation.
The reported 50% reduction in token costs lacks a stated baseline, exact measurement window and calculation method in the supplied material. It is also unclear how many engineers use each harness, how often managers approve spending above tier limits, or how AI Guard and risk scores perform in practice. The account is an interview-based report, not a technical audit of Uber’s systems.
cost management tools for AI development
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Uber Measures Agent Use
The interview does not announce a rollout date or a new product launch. The next developments to watch are whether Uber shares more detail on agent use, code-review outcomes and spending controls, and whether its reported cost reductions are sustained as usage changes. For now, the available account describes an evolving internal system rather than a completed assessment of AI’s effect on engineering productivity.
security middleware for AI systems
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did The Pragmatic Engineer report about Uber’s AI infrastructure?
It described an internal Model Gateway, the AIFX layer for coding harnesses, spending controls and tools that support code review. The details came from an interview with Uber Distinguished Engineer Uday Kiran Medisetty.
Which coding tools does Uber support through AIFX?
The interview says Uber currently supports Claude Code, Codex and OpenCode. AIFX is designed to carry settings, skills and plugins when engineers switch between them.
Does AI replace human pull-request review at Uber?
No. The report says engineers still review all pull requests. AI review, risk scores and supporting screenshots or videos are used to inform that process.
How much has Uber reduced AI token costs?
The interview reports a reduction of about 50% in recent months. It does not specify the baseline, precise time window or calculation method, so the figure should be treated as a company-reported claim.
What remains unknown about the impact of Uber’s AI tools?
The supplied account does not quantify changes in developer productivity, defect rates or review time, nor does it give adoption rates for the tools. Those outcomes cannot be established from the interview alone.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
