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
The Pragmatic Engineer published an interview with Uber Distinguished Engineer Uday Kiran Medisetty about the company’s developer infrastructure and its adoption of AI agents. Uber’s approach builds on existing internal systems, including a model gateway, tools for switching AI harnesses and human-reviewed pull requests; the interview also describes cost controls and unresolved code-review bottlenecks.
The Pragmatic Engineer has published an interview with Uday Kiran Medisetty, a longtime Uber engineer who leads much of the company’s agentic AI work, describing how Uber’s internal infrastructure connects AI tools to its code, services and engineering knowledge. The account offers a view of how a large technology company is adapting developer workflows while retaining engineer review of all pull requests.
Medisetty described a toolkit that includes more than 20 internal developer tools, among them DevPods, Code Inbox, Slate and Wisdom, in a broader look at how AI systems are powered. The interview says Uber’s AI infrastructure extends systems built before generative AI, rather than relying only on newly introduced coding agents. The source does not provide a full technical specification for each named tool.
A central component is Model Gateway, developed by Uber’s platform team in 2023 to keep AI applications independent of a single model provider or harness, an approach that reflects broader questions about how AI works. According to the report, it supports models from Anthropic and OpenAI, as well as open models Uber runs, including tools such as Anthropic’s Claude. The gateway also connects to security checks, per-user spending records, audit logs and session traces. Uber’s AI Foundations wrapper, or AIFX, supports Claude Code, Codex and OpenCode and carries settings, skills and plugins across them.
The interview says Uber gives engineers AI spending tiers, with managers able to approve additional monthly spending when needed. A plugin called CostCheck offers suggestions for reducing costs across sessions. The report says Uber has cut token costs by about 50% “the last few months,” but does not specify the exact measurement period or baseline. Engineers still review every pull request; AI review, risk scores and, where applicable, screenshots or videos are used to help them judge changes.
How Uber Is Connecting Agents to Engineering
The account matters because it shows how AI coding tools depend on existing infrastructure as well as on the models themselves. At Uber, a shared gateway provides a way to change providers, monitor spending and apply security checks, while AIFX gives engineers a way to change coding harnesses without losing their working setup. These choices can affect how easily a company can adopt new tools and oversee their use.
The interview also points to a practical limit: human code review remains part of the process, and Medisetty says review has become a bottleneck. AI-generated review and risk indicators may help engineers allocate attention, but the source does not say that they remove the need for human judgment or report measured improvements in review speed or software quality.
As an affiliate, we earn on qualifying purchases.
Infrastructure Built Before Generative AI
Uber’s internal systems grew in response to needs its engineers said commercial products could not meet at the time. The interview cites uOwn for tracking service ownership, M3DB for handling a reported 500 million metrics per second, Cadence for durable workflows and Morpheus for feature flags and experimentation. The source says Temporal was built on the open-sourced Cadence framework and that Chronosphere was built on M3DB.
The report links this in-house approach to Uber’s operating model, where cities are a basic unit that affects systems from observability to deployment and infrastructure. It says each city is pinned to a cloud region and can fail over to another. This background helps explain why the company developed internal platforms that its AI tools can now use; it does not establish that the same architecture would suit other companies.
““What do developer tools look like inside Uber, today? They are very different from what you would see at most places.””
— The Pragmatic Engineer, introducing the interview with Uday Kiran Medisetty
As an affiliate, we earn on qualifying purchases.
Limits of the Published Account
The interview is a company-insider account, not an independent evaluation of Uber’s AI systems. It does not provide benchmarks showing how the tools affect engineering output, defect rates or review times, nor does it detail the security checks’ performance or the exact rules for approving extra AI spending.
The claimed 50% reduction in token costs lacks a stated baseline and precise time window, so it cannot be compared reliably with another period from the information provided. The source also does not specify when the interview was published, describe the full capabilities of every named internal tool, or say how frequently engineers use each AI harness.
As an affiliate, we earn on qualifying purchases.
How Uber Measures the Changes
The source points to continued work on code review, AI spending and agent integration, but gives no launch dates or announced milestones. It says Uber is examining how productivity should be measured as engineers’ roles change, without providing a finalized metric or results.
Further reporting would be needed to determine whether the tools shorten review cycles, change the rate of defects or produce lasting cost savings. For now, the published account describes Uber’s architecture and policies, while their measurable effects remain unspecified.
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 the company’s internal developer tools and how they support AI agents.
What does Uber’s Model Gateway do?
According to the interview, it lets Uber’s AI applications work across different commercial and in-house models. It also connects to security checks, spending records, audit logs and session traces.
Does AI replace human pull-request review at Uber?
No. The source says engineers still review all pull requests. AI review and other signals, such as risk scores, are used to assist that process.
How does Uber manage AI costs?
The report describes spending tiers for engineers, manager approval for additional monthly spending and a CostCheck plugin that offers cost-reduction suggestions. Its reported cost reduction has no precise baseline or time window.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
