TL;DR
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
A report from The Pragmatic Engineer examines how Uber’s developer infrastructure is adapting to AI agents, drawing on an interview with Distinguished Engineer Uday Kiran Medisetty. Uber’s tools include a model gateway, a layer for switching AI coding harnesses, and systems for reviewing code and controlling spending; engineers still review every pull request.
The Pragmatic Engineer has published a report on how Uber connects AI coding tools to its internal engineering systems, based on an interview with Distinguished Engineer Uday Kiran Medisetty. The account describes infrastructure built both before and after generative AI became widely available, including tools for model access, code review, and engineering knowledge. It offers a view of how a large software organization is adapting its development workflow while keeping engineers responsible for reviewing every pull request.
Uber’s AI infrastructure includes a Model Gateway that lets the company use models from Anthropic, OpenAI and other providers, as well as open models it runs itself. According to the report, Uber’s platform team decided in 2023 that AI-built systems should not be tied to a particular model or harness. The gateway also supports security checks, spending records, audit logs and session traces.
A separate layer called AIFX lets engineers switch among coding harnesses, including Claude Code, Codex and OpenCode, while carrying over settings, skills and plugins. The report also describes AI Guard, security middleware that checks requests and responses and tracks latency. Uber allocates engineers spending tiers for AI use; managers can approve additional spending when an employee reaches a limit. A plugin called CostCheck offers suggestions for managing costs across sessions.
AI has not replaced the human code-review requirement, the report says. Uber runs AI code review before engineers assess changes and has built an internal review tool to help them judge the time and attention a change may require. The account also describes tools including DevPods, Code Inbox, Slate and Wisdom, among more than 20 internal tools used by engineers. The supplied material does not explain each tool’s function in detail.
AI Tools Within Uber’s Review Process
The report matters because it shows that deploying AI coding tools inside a large company involves more than choosing a model. Uber’s described setup combines model choice, security checks, cost controls and audit records with existing developer systems. That infrastructure gives the company ways to change providers and tools without making each engineer’s workflow dependent on one vendor.
Code review is a practical constraint in the account: Medisetty says it has become a bottleneck, while the company continues to require engineer review of all pull requests. AI-assisted checks and risk scoring are presented as ways to help reviewers allocate attention, not as a replacement for their responsibility. The report does not provide independent measurements showing how much these tools have changed review times or overall engineering output.
AI code review tools for developers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Infrastructure Built Before Generative AI
The report places Uber’s current AI systems within a longer history of in-house developer infrastructure. Medisetty joined Uber in 2015, when the company had two services, according to the account. By 2019, it had thousands; the report says the service count in 2026 is about the same as in 2019, despite a business it describes as roughly 10 times larger. These figures are presented in the interview and report, rather than as independently verified measurements.
Uber developed internal systems where the report says existing products could not meet its needs. Examples include M3DB for handling metrics, the durable workflow engine Cadence, and Morpheus for feature flags and experimentation. The report also explains that Uber treats cities as a core organizing concept across systems such as deployments, observability and service sharding. That history helps explain why AI tools now need to work with a broad, company-specific engineering environment rather than only with code editors.
“Code review has become a big bottleneck inside the company.”
— Uday Kiran Medisetty, Uber Distinguished Engineer, as characterized in The Pragmatic Engineer report
Limits of the Published Account
The supplied report excerpt does not give detailed descriptions of all the internal tools named, nor does it provide quantified results for changes in review speed, developer productivity or code quality. It says Uber reduced token costs by about 50% in recent months, but the excerpt does not specify the exact measurement window, comparison baseline or calculation method. That figure should be read as a reported company result, not as an independently audited finding.
The account also does not establish how widely each tool is used across Uber, how risk scores are calculated, or what safeguards determine when managers approve spending above an engineer’s allocated tier. The report describes the company’s approach; it does not independently test the tools or compare their outcomes with other organizations.
How Uber’s Workflow Develops
The report points to continued development of Uber’s AI engineering stack, but the supplied material does not announce a launch date, new product release or formal next milestone. The company’s stated approach is to support multiple models and harnesses while applying security checks, tracking costs and retaining engineer review of pull requests.
Further evidence about the impact of the tools would require more detail on adoption, review times, spending and code-quality outcomes. The Pragmatic Engineer’s interview and episode materials provide the account currently available; whether Uber will publish additional measurements or expand the systems described remains unclear.
Key Questions
What is the news about Uber’s AI infrastructure?
The Pragmatic Engineer published an interview-based report describing Uber’s internal developer tools and systems for connecting AI coding assistants with company models, code and engineering workflows.
Does Uber let AI approve code changes without engineers?
No. The report says engineers still review all pull requests. AI review runs before the human review and is part of a process intended to help reviewers assess changes.
Which AI coding tools does Uber support?
The report says Uber supports Claude Code, Codex and OpenCode through AIFX, which carries settings, skills and plugins when engineers switch harnesses.
How does Uber manage AI model choice and spending?
Uber’s Model Gateway supports models from multiple vendors and open models the company runs. Engineers receive spending tiers, and managers can approve additional spending after an employee reaches a limit.
Has Uber proved that these tools increase productivity?
The supplied report does not provide quantified, independently verified productivity results. It describes the tools and reports a roughly 50% reduction in token costs over recent months, without specifying the measurement window or comparison baseline in the excerpt.
Source: rss
Evergreen bestsellers Picks
bestsellers
As an affiliate, we earn on qualifying purchases.
