# Agentic SDLC at Uber — Uday Kiran Medisetty & Adam Huda, Uber

## Executive summary

Uber has implemented a comprehensive 'Software Factory' powered by agentic AI to revolutionize its SDLC. Key infrastructure investments include a centralized Model Gateway for security and latency control, an MCP Gateway for token optimization, and a Context Graph that unifies knowledge across disparate systems. These building blocks enable autonomous agents (like Minion) to accelerate development from idea generation (via Cortana) through drafting PRs, validating changes in inner loops (e.g., comparing simulator screenshots against Figma specs), and managing maintenance tasks.

## Key takeaways

- Agentic Adoption Metrics: Over the last year, agentic AI led to more than 70% of Uber's Pull Requests (PRs) originating from local or cloud agents, resulting in double the lines of code per engineer year-over-year.
- Model Gateway Performance: The Model Gateway handles all internal model calls through a single endpoint, ensuring PII redaction for over 20 types and running five specialized safety models under a strict 100 millisecond guardrail budget. It processes over 100 million requests daily across 800+ projects.
- Token Optimization: The MCP Gateway, combined with projecting MCPS into CLI patterns and using Omni MCP, cut fleetwide token usage by more than 40%.

## Technical details

- Model Gateway (Security & Latency): All model calls pass through a single OpenAI/Anthropic compatible endpoint. It enforces identity via Spire, redacts 20+ PII types, and runs five specialized safety models, all while maintaining sub-100ms latency.
- MCP Gateway (Tooling): The MCP gateway provides a single entry point for internal APIs and SaaS tools (Google, Slack, Jira). It uses an automated crawler to project these into MCPS, simplifying tool access and managing token exchange. Omni MCP allows discovery and invocation of any MCPS within the gateway.
- Dev Environment Scaling: Uber transitioned from traditional DevPods to pre-provisioned Kubernetes balloon pods for agents, allowing quick, isolated environments with snapshotted repositories and built search indexes. A 'mega dev port' now supports autonomous coding across multiple repositories.
- Skills Marketplace: A managed marketplace hosts 2,500 skills (core and domain-specific). It enforces quality via lint checks and automated reviews, running over 20,000 executions daily. Skills can be autoinstalled based on engineer personas.
- Context Graph: A unified context graph consolidates information from 20-30 scattered systems (design docs, Jira, incident bugs) into a single source with 150 unique node and edge types and 40 million entries. This drastically improves agent ability to find basic context compared to using multiple separate skills.
- Agentic SDLC Workflow: The process begins with Cortana (AI assistant) for idea validation in Slack/web interfaces. The Minion agent then drafts a PR, deliberately stopping short of CI. Validation shifts from the outer loop to an inner loop, incorporating visual checks (comparing simulator screenshots to Figma specs) and self-healing CI mechanisms before reaching the build queue.

## Practical implications

- The bottleneck in software development is shifting from implementation capacity to the strategic decision of whether a feature should be built at all.
- Autonomous agents are now capable of handling complex maintenance tasks (e.g., feature flag cleanup) via managed loops, ensuring controlled and predictable code changes.
- By moving validation checks into an 'inner loop' before CI, teams can significantly reduce load on build infrastructure while improving quality.

## Topics

Agentic AI, Software Development Life Cycle (SDLC), DevOps, AI Infrastructure, Monorepos, Minion agent, Cortana

Source: https://www.youtube.com/watch?v=17-YSUHo6Lk