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’s developer infrastructure is being adapted for AI agents. Uber’s account covers a shared model gateway, tools for switching AI coding harnesses and controlling spend, and human-led code review supported by automated checks.
Uber’s internal AI infrastructure connects coding agents with company code, services and engineering knowledge, according to Distinguished Engineer Uday Kiran Medisetty in an interview published by The Pragmatic Engineer. The report describes a model gateway, tools for using different coding-agent harnesses and controls on AI spending, alongside a review process in which engineers still review every pull request.
Uber’s platform team decided in 2023 that AI-built tools should be independent of any single model or harness, the report says. The company built a Model Gateway to route requests to models from Anthropic, OpenAI and other providers, as well as open models Uber runs itself. The gateway also supports security checks, latency monitoring, per-user spend tracking and audit logs and session traces.
A separate tool called AIFX, short for AI Foundations, sits above coding harnesses including Claude Code, Codex and OpenCode. The report says it lets engineers switch harnesses while keeping settings, skills and plugins. Uber also uses spend tiers: engineers have an allocated budget, and a manager can approve additional spending for that month. A plugin called CostCheck offers suggestions across sessions to help reduce costs.
For code review, the interview describes automated checks and risk scores intended to help engineers decide how to assess changes. AI review runs before human review, and screenshots or videos can be attached when relevant. Medisetty says code review has become a bottleneck; the source material says all pull requests are still reviewed by engineers. The supplied report excerpt ends during its account of this review tooling, so it does not provide a complete description of how risk scores are calculated or what follows each score.
How Uber Governs AI Coding
The account shows how an established technology company is adding AI coding tools to existing engineering systems rather than treating them as stand-alone assistants. A shared gateway gives Uber a way to route work across providers while applying security checks, spending controls and audit logging. The approach may also reduce dependence on a single model vendor, although the report does not compare performance or costs across providers.
The human review requirement matters because AI-generated code still enters a process where engineers must evaluate changes. Automated checks and risk scores are presented as ways to focus that work, not as replacements for it. At a company with a large service ecosystem, the challenge is not only generating code but also managing its reliability, ownership and operational risks.
Uber’s account of costs adds a practical constraint: broad agent use can generate substantial model usage, so the company combines allocated budgets with manager approval for additional spending. The report says Uber reduced token costs by about 50% in recent months, but the supplied material gives no precise time window, baseline or methodology. That figure should be read as the company’s reported result, not as an independently established benchmark.
As an affiliate, we earn on qualifying purchases.
Infrastructure Built Before Agents
Uber’s AI tools sit on top of an infrastructure built for the company’s earlier scale challenges. Medisetty joined in 2015, when the company had two services, according to the report. Uber had thousands by 2019. The report says its service count in 2026 is expected to be around the same as in 2019, despite the business being about ten times larger; it does not provide a precise service count for either year in the supplied excerpt.
The interview describes several systems Uber built in-house when available products could not meet its needs. These include uOwn for service ownership, M3DB for metrics, Cadence for durable workflows and Morpheus for feature flags and experimentation. The report says Chronosphere grew from open-source M3DB and Temporal from open-source Cadence.
Uber’s city-based operations also shaped its infrastructure: cities are treated as a core concept across systems, from observability and deployments to data sharding. The report says each city is assigned to a cloud region and can fail over to another. This history helps explain why the company’s agent tools need to connect with internal services and engineering context, not just source code.
“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 monitoring software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Details the Interview Leaves Open
The supplied material does not specify the publication date, the number of engineers using these tools, or how much code is generated or reviewed with AI. It also does not explain the Model Gateway’s technical design, the criteria used for code risk scores, or how the reported roughly 50% token-cost reduction was measured. The excerpt cuts off during the code-review discussion, leaving further details about the workflow unavailable.
The descriptions of internal tools and their effects are based on Uber’s account in the interview. The source does not provide independent verification, comparative productivity data or evidence that the tools have reduced review times. It is also unclear how often managers approve additional monthly spending or how Uber measures the quality of AI-assisted changes.
cost management tools for AI development
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
More Evidence on Productivity
The Pragmatic Engineer page points readers to the episode transcript and timestamps, as well as versions on YouTube, Apple and Spotify. The supplied material does not announce a new launch date or a forthcoming Uber milestone. The main open question is how the described systems perform in practice: whether they reduce review delays, maintain code quality and keep AI spending within useful limits. More detailed usage, cost and productivity measures would help establish those outcomes.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the news about Uber’s AI infrastructure?
The Pragmatic Engineer published an interview with Uber Distinguished Engineer Uday Kiran Medisetty about internal tools that connect AI coding agents to Uber’s models, code and engineering workflows.
Which AI models and coding tools does Uber support?
The report says Uber’s Model Gateway routes requests to Anthropic, OpenAI and other providers, including open models run by Uber. AIFX supports Claude Code, Codex and OpenCode as coding harnesses.
Does AI review replace engineers at Uber?
No. The source says engineers still review every pull request. Automated AI review and risk scores are used to inform that process.
How does Uber control AI spending?
Engineers receive a spend tier, and managers can approve additional spending for a month. The company also uses CostCheck to suggest ways to reduce usage costs, according to the interview.
Has Uber proved that these tools improve productivity?
The supplied report describes the tools and calls code review a bottleneck, but it does not provide independent productivity results, review-time comparisons or a full methodology for its reported token-cost reduction.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
