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 report by The Pragmatic Engineer details how Uber is adapting its existing developer infrastructure for AI coding agents. The company uses a shared model gateway, supports several coding harnesses and keeps human engineers responsible for reviewing pull requests; the report says token costs have fallen about 50% in recent months.
The Pragmatic Engineer has published an inside look at Uber’s AI infrastructure and developer tools, based on an interview with Distinguished Engineer Uday Kiran Medisetty. The report describes how Uber connects coding agents to its models, code and engineering knowledge while keeping human engineers responsible for reviewing pull requests.
Uber’s platform team decided in 2023 that AI systems built at the company should be independent of any one model or coding harness, according to the report, an approach also explored in Anthropic’s AI agent workflows. Its Model Gateway provides a common route to models from Anthropic and OpenAI, as well as open models Uber runs itself. The company can change providers without building each internal tool around a single vendor.
The gateway also supports controls and monitoring. Uber’s security team runs AI Guard middleware to check requests and responses and measure latency. The platform records spending by user and keeps audit logs and session traces, the report says. Uber’s AI Foundations wrapper, or AIFX, supports Claude Code, Codex and OpenCode, carrying settings, skills and plugins between them, while Anthropic’s Claude Frontier Academy focuses on training people to work with AI.
Engineers receive an AI spending tier; managers can approve additional monthly spending when needed. A plugin called CostCheck offers advice on reducing costs across sessions. The report says Uber has cut token costs by about 50% in recent months, but does not specify the measurement period, starting cost or methodology behind that figure.
How Uber Controls AI Coding
The infrastructure reflects a practical challenge for companies adopting coding agents: giving engineers access to multiple models and tools while managing security, cost and oversight. A shared gateway can make it easier to change model providers, while centralized checks and logs give the company a way to monitor how those systems are used.
Uber’s approach also shows that introducing agents does not, in the account given, remove human responsibility for code quality. The report says engineers still review every pull request. AI review, risk scores and supporting materials such as feature screenshots or videos are intended to help reviewers judge how much attention a change needs. Medisetty describes code review as a growing bottleneck, making the review workflow a central part of the company’s AI strategy rather than an afterthought.
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 building internal engineering infrastructure. Medisetty joined Uber in 2015, when the company had two services, according to the account. The number grew to thousands by 2019. The report says that in 2026 Uber expects to have about as many services as in 2019, despite a business roughly ten times larger than it was then.
Uber developed tools such as M3DB for metrics and Cadence for durable workflows because existing products could not meet its needs at the time, the report says. Its systems also account for the company’s city-based operations: cities affect areas including observability, deployments and service sharding, and each city is assigned to a cloud region that can fail over to another. That established platform is the foundation the company is now connecting to AI agents.
““everything built with AI needs to be model- and harness-agnostic.””
— The Pragmatic Engineer, summarizing Uday Kiran Medisetty’s account
developer infrastructure monitoring software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits of the Reported Results
The published account does not give independent measurements for the reported 50% reduction in token costs, including the baseline, exact timeframe or whether the figure covers all engineers and workloads. It also does not quantify how much the tools have changed development speed, review time or software quality.
The source describes Uber’s internal systems through an interview and company experience; it is not an independent audit of their effectiveness. The report excerpt also does not provide detailed results for every tool it mentions, so the broader impact of background agents and risk scoring remains unclear.
As an affiliate, we earn on qualifying purchases.
Further Evidence on Agent Impact
The report does not announce a specific rollout date or upcoming product milestone. The next useful evidence would be more detail from Uber on performance and cost measurement: how it calculates token savings, whether review bottlenecks are easing, and what safeguards apply as agents take on more coding work. The Pragmatic Engineer’s interview offers a snapshot of the systems in use, but their longer-term effects are not established in the material provided.
cost management tools for AI development
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did The Pragmatic Engineer report?
It described Uber’s internal developer infrastructure and how the company connects AI models and coding agents to engineering workflows, drawing on an interview with Uday Kiran Medisetty.
Which AI models and coding tools does Uber support?
The Model Gateway connects to models from Anthropic and OpenAI, along with open models Uber runs. AIFX supports the coding harnesses Claude Code, Codex and OpenCode, according to the report.
Do AI agents approve Uber pull requests?
No. The report says engineers still review all pull requests. AI review, risk scores and other context are used to help inform the human review process.
How much has Uber reduced token costs?
The report says token costs fell by about 50% in recent months. It does not give the baseline, exact timeframe or calculation method, so the figure cannot be assessed in more detail from the source alone.
Source: rss
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
