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 report and interview with Uber Distinguished Engineer Uday Kiran Medisetty details how the company’s existing developer infrastructure is being extended for AI agents. Uber’s systems include a model gateway, tools for switching AI coding harnesses, spending controls and human review workflows; the report does not provide independent measurements of productivity gains.
The Pragmatic Engineer has published an inside look at Uber’s developer infrastructure for AI-assisted coding, based on an interview with Distinguished Engineer Uday Kiran Medisetty. The report describes how the company connects coding agents to models, code and internal engineering knowledge, while retaining human review of pull requests.
Uber’s AI setup builds on internal tools developed before generative AI became widely used. The report describes a Model Gateway introduced in 2023 to let teams use different models and switch among vendors, as well as open models Uber runs itself. A security middleware called AI Guard checks requests and responses and tracks latency; the gateway also supports per-user spending records, audit logs and session traces.
For coding agents, Uber uses AIFX, a wrapper intended to carry settings, skills and plugins across supported harnesses, including Claude Code, Codex and OpenCode. The company also allocates engineers spending tiers for AI use. According to the report, a manager can approve additional spending after an engineer reaches a limit, and a plugin called CostCheck offers advice on reducing costs across sessions.
The interview also covers tools developed for engineers’ day-to-day work, including DevPods, Code Inbox, Slate and Wisdom. Uber has built a code-review system to help engineers judge how much attention a change needs. The report says AI review, feature screenshots or videos where relevant, and risk scores can be available before a human reviews a pull request. Engineers still review all PRs, Medisetty said, while identifying code review as a growing bottleneck.
How Uber Governs AI Coding
The report offers a practical account of how a large engineering organization is adapting existing infrastructure rather than treating coding agents as standalone tools. A shared gateway gives Uber a way to manage model choice, security checks and usage visibility across different providers. That can matter as models and coding products change, since a common layer can make it easier to switch systems without rebuilding every internal integration.
Spending controls and human review address two concerns that accompany broader agent use: unpredictable model costs and code changes that still need engineering judgment. Uber’s approach, as described in the interview, allows managers to approve higher usage when needed, but does not remove review responsibility from engineers. The report outlines the company’s practices; it does not establish that the tools have increased productivity or reduced defects.
As an affiliate, we earn on qualifying purchases.
Infrastructure Built Before Agents
Uber’s internal development systems grew alongside a rapid expansion in services. The report says the company had two services in 2015 when Medisetty joined, and thousands by 2019. It estimates that in 2026 the service count will be about the same as in 2019, despite the business being roughly 10 times larger than it was then. The 2026 figure is presented as an estimate, not a final audited count.
The report says Uber built tools such as uOwn for service ownership, M3DB for metrics, Cadence for durable workflows and Morpheus for feature flags and experimentation when available commercial products did not meet its needs or were not available. It also describes cities as a basic operating concept that affects systems from deployments to observability and infrastructure. That history helps explain why Uber had internal platforms that could later be adapted to connect AI agents with company systems.
““Cities are a basic primitive.””
— Uday Kiran Medisetty, Uber Distinguished Engineer, as quoted in The Pragmatic Engineer report
developer infrastructure management software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Productivity Results Not Quantified
The report describes Uber’s tools and operating choices but provides no independently measured productivity results, code-quality comparison or detailed evaluation of how often agents produce changes that require substantial correction. It also does not give a full account of which engineering tasks are handled by agents or how adoption varies across teams.
The source says Uber has reduced token costs by about 50% in recent months, but the supplied material does not specify the exact measurement window, baseline, or whether that figure is independently verified. It should be treated as a reported company figure, not a general measure of AI savings. More detail on risk-score methods and the practical effect of the review system is also not provided.
As an affiliate, we earn on qualifying purchases.
How Agent Workflows Develop
The report does not announce a launch date or a new public product. Its next developments are internal: Uber’s teams will continue using and refining the model gateway, harness wrapper, spending controls and review tools as coding agents and model options change. The company’s ability to switch among providers is designed to accommodate that change, according to the report.
For readers assessing the approach, the key evidence to watch for is more detail on usage, costs and engineering outcomes: how the systems affect review time, the quality of code changes and the amount of human rework. The source does not say when Uber will publish such measurements or whether it plans to disclose them.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the main development in the report?
The Pragmatic Engineer published an interview with Uber Distinguished Engineer Uday Kiran Medisetty describing internal tools and infrastructure for AI-assisted software development.
Which AI models and coding tools does Uber support?
The report says Uber’s Model Gateway supports models from multiple vendors and open models run by Uber. Its AIFX wrapper supports Claude Code, Codex and OpenCode as coding harnesses.
Do AI agents replace human code review at Uber?
No. According to the report, engineers still review every pull request. Automated review, screenshots or videos where relevant, and risk scores are among the information available to help reviewers.
How does Uber control AI spending?
Uber assigns engineers spending tiers, and managers can approve additional spending after a limit is reached, according to the report. Its CostCheck plugin gives engineers suggestions about their cost profile.
Has Uber shown that these tools improve productivity?
The report explains the tools and Medisetty’s account of their use, but does not provide independent productivity measurements or a quantified comparison of engineering outcomes.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
