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 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 coding harnesses, spend controls and AI-assisted code review; engineers still review every pull request.
Uber’s internal AI infrastructure connects coding agents to company models, code and engineering workflows, according to a new interview with Uday Kiran Medisetty, a Distinguished Engineer at the company. In the Pragmatic Engineer episode, Medisetty describes systems for routing model requests, controlling costs and helping engineers review code, offering a view of how a large technology company is adapting developer tools built before generative AI.
Uber’s Model Gateway, introduced by the platform team in 2023, is designed to let internal AI products use different models and harnesses rather than depend on a single provider. The report says it supports models from Anthropic and OpenAI, as well as open models Uber runs. Security middleware called AI Guard checks requests and responses and tracks latency. Uber also records spending by user and keeps audit logs and session traces.
A separate layer, AIFX, helps engineers move among Claude Code, Codex and OpenCode while retaining settings, skills and plugins. The report describes a tiered spending policy: engineers receive an allocation, and a manager can approve additional spending after the limit is reached. A plugin called CostCheck offers suggestions across sessions. The source says Uber reduced token costs by about 50% in recent months, but does not specify the measurement period or comparison baseline.
AI is also being added to code review, which Medisetty describes as a bottleneck. According to the report, AI review, risk scores and supporting screenshots or videos can be available before a human reviews a change. Engineers still review all pull requests; the tools are intended to help them judge how much attention each change needs, not remove human review.
How Uber Controls AI Coding
The systems show that adopting coding agents at a large company involves more than selecting a model. Uber’s approach puts model choice, security checks, usage records and spending controls around the tools engineers use, while preserving a human review step. That framework can let teams try different models without rebuilding every internal integration when a provider or coding harness changes.
The review workflow also addresses a practical limit on AI-assisted development: faster code generation can leave engineers with more changes to inspect. Risk scores and automated checks may help direct reviewer attention, but the interview does not establish how much review time they save or whether they improve defect rates.
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 AI
The interview presents Uber’s AI tools as an extension of a longer history of building developer infrastructure in-house. Medisetty joined in 2015, when the company had two services, according to the report. Its service count rose to thousands by 2019; the report says it is expected to be around that level in 2026, despite the business being about ten times larger than in 2019. The 2026 figure is presented as an estimate, not a confirmed current count.
The source cites systems developed for needs that available products could not then meet: uOwn for service ownership, M3DB for metrics at high volume, Cadence for durable workflows and Morpheus for feature flags and experimentation. Uber’s city-based operations also shape its infrastructure, from observability and deployments to regional placement. That history helps explain why the company’s AI tooling is described as integrated with internal systems rather than as a standalone assistant.
“The source says code review has become a big bottleneck inside the company.”
— Uday Kiran Medisetty, Uber Distinguished Engineer, as described in The Pragmatic Engineer interview
As an affiliate, we earn on qualifying purchases.
What Uber Has Not Disclosed
The interview excerpt does not provide independent measurements of how Uber’s AI tools affect engineering speed, code quality or incidents. It also does not define the baseline or exact period behind the reported 50% reduction in token costs, so that figure should be treated as the source’s account rather than a directly comparable efficiency measure.
Some details of the code-review system are missing from the supplied material, including how its risk scores are calculated and how often they change a reviewer’s decision. The report also does not give current figures for AI usage, model spending or the number of engineers using each tool.
cost management tools for AI development
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How the Tooling Develops
The episode serves as a description of Uber’s current approach, not an announcement of a launch date or a new product rollout. The source material gives no timetable for further changes. The next useful evidence will be whether Uber publishes clearer measurements of cost, review workload and engineering outcomes as it expands agent use. For now, the stated policy remains that AI can assist with code review, but engineers review every pull request.
security middleware for AI applications
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is Uber’s Model Gateway?
It is an internal platform for routing AI work across supported models. The report says it supports Anthropic and OpenAI models and open models run by Uber, with security checks, spending records and audit traces around requests.
Can Uber engineers use different coding agents?
Yes. The report says Uber supports Claude Code, Codex and OpenCode. Its AIFX layer carries settings, skills and plugins when an engineer switches harnesses.
Does AI replace human code review at Uber?
No. According to the report, AI review and risk information can help engineers assess changes, but all pull requests are still reviewed by engineers.
How does Uber manage AI spending?
Engineers receive a spending tier, and managers can approve additional monthly spending after the allocation is reached. CostCheck also offers suggestions for reducing usage costs.
How much has Uber cut token costs?
The source reports a reduction of about 50% in recent months. It does not state the comparison baseline or precisely define the period, so the figure cannot be interpreted as a fully specified cost comparison.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
