Enterprise Internal Deployments
Companies publishing detailed data about internal AI coding deployments at scale. These reveal production adoption patterns that commercial vendor data does not capture.
| Company | Level | Key Metric | Detail |
|---|---|---|---|
| L9-L10 | 30%+ of new code AI-generated | CEO Sundar Pichai (earnings calls, up from 25% in Q3 2024). Internal "Goose" tool on Gemini leverages 25 years of proprietary engineering knowledge. Plans to hire more engineers because "the opportunity space is expanding". | |
| Uber | L10 | ~70% of committed code machine-generated | ~95% of engineers use AI monthly. Multi-agent system: Validator, AutoCover, uReview (90%+ of ~65,000 weekly diffs). AI coding agent: <1% to ~8% of code changes in months. 21,000 developer hours saved. Specialised agents outperform general-purpose. |
| Stripe | L10 (partial L11) | 1,300+ PRs/week, zero human code | Minions system. Fork of Goose. ~500 MCP tools. Blueprints architecture. Some system-initiated triggers. |
| Ramp | L10 | 50%+ of merged PRs | Inspect system. OpenCode on Modal. Full dev env per sandbox. 80% of Inspect self-written. |
| Coinbase | L10 | 25+ hours/week saved per automation | Cloudbot, RAPID-D, qa-ai-agent, LangGraph enterprise agents. Paved roads: 12+ weeks to <1 week for new agents. x402 protocol. |
| Shopify | L5-L8 | AI usage in performance reviews | CEO Tobi Lutke memo (April 2025): "reflexive AI usage is now a baseline expectation." Teams must prove why AI can't do a task before requesting headcount. Some employees achieving "100x expected output." Provides Copilot, Cursor, Claude Code, internal chat tool. |
| Microsoft | L5-L9 | 20-30% of codebase AI-generated | CoreAI Platform and Tools division formed January 2025, merging Dev Division, AI Platform, and CTO teams. Some teams use competitor tools alongside Copilot. |
| Netflix | L5-L9 | 150,000 internal support tickets/year via AI | Productivity Assistant team partnered with Anthropic. Universal Search reclaimed 416 hours/employee annually. Hiring AI PMs at $240K-$700K. |
| Meta | L4-L5 | 20%+ suggestion acceptance | CodeCompose: 6.7B parameter model fine-tuned on internal code (including Hack language). Thousands of employees accepting suggestions weekly. |
| Block | L6-L10 | 5,000 weekly Goose users | Open-sourced Goose agent (29,400+ stars). Stripe forked Goose for Minions. Contributed to AAIF. |
| StrongDM | L11 | No human writes or reviews code | 3-person AI team. $1,000+/day tokens per engineer. Digital Twin Universe. Open-sourced Attractor + cxdb. |
| MiniMax | L11 (self-reported) | 80% of code, 30% of tasks autonomous | M2.5 model used internally. Not independently verified. |