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 by The Pragmatic Engineer describes how Uber is adapting its existing developer infrastructure for AI agents, including a model gateway, tools for switching coding assistants and systems for code review. Uber Distinguished Engineer Uday Kiran Medisetty also discusses spending controls and productivity measurement; the report does not provide independent evidence that these tools have increased productivity.
The Pragmatic Engineer has published a report and interview detailing how Uber is adapting its internal developer infrastructure for AI-assisted engineering, including a shared model gateway, tools for coding agents and changes to code review. Uber Distinguished Engineer Uday Kiran Medisetty, who leads much of the company’s work on agentic AI, described the systems and practices in the episode.
Uber’s approach builds on infrastructure developed before generative AI became widely used by software engineers. The report says the company created internal systems for service ownership, metrics, workflows and experimentation when available commercial products did not meet its needs. It describes those systems as part of the foundation now used to connect AI tools with Uber’s code, internal services and engineering knowledge.
A central component is Model Gateway, which Uber’s platform team began building in 2023. The approach is also relevant to AI-native engineering ecosystems. According to the report, the gateway is designed to let teams use models from Anthropic, OpenAI and other vendors, as well as models Uber runs itself, without tying applications to one model provider. Security middleware called AI Guard checks requests and responses, while the platform tracks spending by user and records audit logs and session traces, priorities also explored in AI security efforts.
The report also describes AIFX, a wrapper supporting Claude Code, Codex and OpenCode that carries settings, skills and plugins across coding harnesses, a development closely related to multi-agent AI systems. Uber allocates engineers spending tiers, with managers able to approve additional monthly spend. A plugin called CostCheck offers suggestions for reducing costs. Medisetty said code review remains a bottleneck: AI review runs before human review, and reviewers may receive supporting material such as screenshots, videos and risk scores. The report says engineers still review all pull requests.
How Uber Connects Agents to Engineering
The report matters because it offers a view of how a large technology company is incorporating AI into established engineering workflows rather than treating coding assistants as standalone products. A common gateway can give Uber a way to change model providers, apply security checks and track costs across tools. That flexibility may be useful as models and coding products change, though the source does not quantify its effect on software delivery.
Human review remains part of the process described. AI-generated code and automated review may change what engineers inspect and how they allocate attention, but Uber’s account does not establish that review time has fallen or that the tools have improved code quality. The company’s spend tiers also show a practical tension: enabling engineers to use AI while monitoring usage and controlling costs.
Products Worth Considering

Jackery Explorer 1000 v2 Portable Power Station, 1070Wh, 1500W

Amazon Kindle Paperwhite Signature Edition (12th Generation, 2024), 32 GB, Metallic Raspberry

LG UltraGear 25G590B-B 25-Inch Full HD 1000Hz Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Infrastructure Built Before AI Coding
The interview places the AI systems within a longer history of Uber’s internal tooling. Medisetty joined the company in 2015, when, according to the report, Uber had two services. Its service count had reached thousands by 2019; the source says it is expected to be around that level again in 2026, despite the business being about ten times larger than in 2019. These figures are presented in the report, not independently verified here.
Examples of earlier systems include uOwn for identifying service owners, M3DB for handling metrics, Cadence for durable workflows and Morpheus for feature flags and experimentation. The report says Uber built these tools because suitable products were unavailable or could not handle the company’s requirements at the time. It also notes that Uber’s systems account for cities as a core operational unit, including in observability, deployments and infrastructure decisions. That legacy helps explain why the company’s AI effort includes connecting agents to internal systems, not only selecting a model.
“All PRs are still reviewed by engineers.”
— The Pragmatic Engineer, describing Uday Kiran Medisetty’s account
Productivity and Cost Claims Need Data
The source describes Uber’s systems through an interview and company-specific account; it does not include independent performance data, detailed methodology or comparative results showing how the tools affect productivity, reliability or review times. It says Uber reduced token costs by about 50% in recent months, but the supplied material does not give the precise measurement window, baseline or calculation method. That figure should be read as a reported company result, not a verified industry comparison.
The report also does not provide full technical details about how AI Guard evaluates requests, how risk scores are calculated, or what data is used to measure changes in engineer productivity. The scope of agent access to internal systems and the outcomes of human reviews are not specified in the supplied material.
More Detail on Tools and Measurement
The Pragmatic Engineer’s report points readers to a podcast episode and transcript discussing Uber’s internal developer tools, AI agents, review workflows, token spending and productivity measurement. The supplied source does not announce a product launch, a rollout schedule or a new company policy. Further public detail on adoption, spending trends and measured engineering outcomes would help clarify how the systems are performing.
Key Questions
What did The Pragmatic Engineer report about Uber?
It published an interview and account of Uber’s internal developer infrastructure, focusing on how existing systems and newer tools support AI-assisted engineering.
What is Uber’s Model Gateway?
According to the report, it is a shared layer that lets Uber applications use models from multiple vendors and models operated by Uber. It also supports security checks, spending tracking and audit records.
Does AI now review Uber’s code instead of engineers?
No. The report says AI review runs before human review, but engineers still review all pull requests.
How does Uber manage AI spending?
The report says engineers receive spending tiers, and managers can approve additional monthly spend. Uber also uses a plugin called CostCheck to suggest ways to reduce costs.
Has Uber shown that these tools improve productivity?
The supplied report describes Uber’s tools and approach but does not provide independent evidence or detailed measurements establishing a productivity gain.
Source: rss
