TL;DR
Get business pricing on networking and server 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 is adapting its internal developer infrastructure for AI coding agents. Uber uses a model gateway, tools for switching coding harnesses and managing token spend, and AI-assisted code review, while engineers still review every pull request.
Uber’s internal engineering platform now connects coding agents to company code, services and documentation through tools built on top of infrastructure developed before generative AI, according to a Pragmatic Engineer interview with Distinguished Engineer Uday Kiran Medisetty. The discussion explains how Uber handles model access, agent settings, token costs and code review as its engineers adopt AI-assisted development.
Uber’s approach includes a Model Gateway designed to keep AI-powered tools independent of any one model or coding harness. The company began that work in 2023, according to the interview. The gateway supports models from Anthropic and OpenAI, as well as open models Uber runs. Uber also uses it for security checks, spending records and visibility into requests through audit logs and session traces.
A separate wrapper called AIFX, short for AI Foundations, lets engineers move among Claude Code, Codex and OpenCode while retaining settings, skills and plugins. The interview describes a spending system that assigns engineers tiers; a manager can approve extra spend for a month when an engineer reaches a limit. Uber also uses a plugin called CostCheck to offer suggestions across sessions. The report says the company reduced token costs by about 50% in recent months, but does not provide the measurement method or exact period.
Code review remains a human responsibility: engineers review all pull requests, while AI review and other tools help them assess changes. The system can provide risk scores and, where relevant, screenshots or videos of a feature before a person reviews the code. Medisetty says review has become a bottleneck at Uber. The interview also discusses background agents and internal tools such as Slate, but the supplied material does not fully describe every tool or its operation.
How Uber Is Governing AI Coding
The account illustrates a practical challenge for companies introducing coding agents: giving them access to useful engineering systems while retaining security controls, cost oversight and human review. Uber’s model gateway provides a common layer for different models, while AIFX reduces the friction of changing the agent interface engineers use. Those choices may make it easier for the company to adopt new models without rebuilding every internal integration.
The interview also highlights that AI adoption does not remove existing engineering bottlenecks. If agents generate or modify code faster, review capacity and risk assessment become more consequential. Uber’s tools are intended to help engineers allocate attention, but the report does not establish how much review time they save or whether they have improved software quality.
As an affiliate, we earn on qualifying purchases.
Infrastructure Built Before Generative AI
Uber’s AI tools sit within a broader history of internal developer infrastructure. Medisetty joined the company in 2015, when it had two services, according to the report. By 2019 it had thousands; the interview says Uber expects to have about the same number in 2026, despite a business roughly ten times larger than in 2019. These figures are presented in the interview, not independently verified in the supplied material.
The report says Uber built systems such as uOwn for service ownership, M3DB for high-volume metrics, Cadence for durable workflows and Morpheus for feature flags and experimentation. The source links some later products and companies to open-source work originating at Uber: Temporal was built on Cadence, and Chronosphere was built on open-sourced M3DB. Uber’s city-based operations also shaped its infrastructure, including how services are observed, deployed and assigned to cloud regions.
That foundation helps explain why the company could add AI capabilities through shared internal platforms rather than treating each coding tool as a standalone product. The interview frames the current systems as an extension of Uber’s long-running practice of building custom engineering tools for its scale and operating needs.
“Code review has become a big bottleneck inside the company.”
— Uday Kiran Medisetty, Uber Distinguished Engineer, as described in The Pragmatic Engineer interview
developer infrastructure management software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
What the Interview Does Not Quantify
The supplied report does not give independent performance data showing how much the tools improve developer productivity, shorten review times or reduce defects. It also does not explain the methodology behind the reported 50% token-cost reduction, or provide a precise time window and comparison baseline.
Some details are incomplete in the source excerpt, including the full operation of Slate, the risk-scoring system and the background agents. The account is based on an interview with an Uber engineer; it does not include independent verification or comments from other employees. It is also unclear how the company measures the balance between faster code production and the time required for human review.
As an affiliate, we earn on qualifying purchases.
How Uber Measures Agent Adoption
The interview does not announce a specific rollout date or next product milestone. The next indicator will be whether Uber shares more detail on agent usage, review outcomes and token spending, including how those figures are measured. For now, the company’s stated approach is to support several models and harnesses, monitor their use and keep engineers responsible for pull-request review.
cost management tools for AI development
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Uber’s Model Gateway?
It is a shared layer for connecting AI-powered tools to models from Anthropic, OpenAI and open-model providers, while supporting security checks, spending tracking and session visibility.
Which coding agents or harnesses does Uber support?
The interview says Uber supports Claude Code, Codex and OpenCode. AIFX carries settings, skills and plugins across those harnesses.
Do AI tools approve Uber’s pull requests?
No. According to the report, engineers still review all pull requests. AI review, risk scores and supporting media help inform that work.
How does Uber control AI costs?
Engineers receive spending tiers, and managers can approve additional monthly spending after a limit is reached. Uber also uses CostCheck to suggest ways to improve an engineer’s cost profile.
Has Uber published proof that these tools improve productivity?
The supplied interview discusses the tools and reports a roughly 50% reduction in token costs in recent months, but it does not provide productivity results or a detailed methodology for the cost figure.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
