DoorDash runs 130K automated tasks a month: inside 8 AI software factories
Pavan_Belagatti · x · 2026-10-20
Port's "software factories" series breaks down how top engineering teams run AI agents across the SDLC:
- DoorDash: in-house cloud-agent platform Flux grew from code review to 130,000 automated engineering tasks a month; its reviewer covers 10,000+ PRs a week across 56 repos and pushes fixes back to PRs
- Ramp: 4 engineers + 1 part-time PM built a background agent that now authors 3 of 4 merged PRs, with the same tool access as a human engineer
- Shopify: a Slack-native agent coauthors 1 in 8 merged PRs; security requires proof-by-exploit and it refuses to count AI-authored lines of code
- Snap: a three-part pipeline (Casper, CodePal, Code Search), with over half of agent investment going to guardrails
- Stripe: one-shot end-to-end agents on a forked Goose, gated by a two-round CI cap and a shared 500-tool internal MCP server
- Uber: six platform layers, an agent registry tracing every action to a person, and an honest account of what AI measurement got wrong
- WorkOS: an issue-to-PR factory in isolated sandboxes that won't open a PR without human approval
More from coding & agent
- Solo builder asks Reddit: how to add client dashboards on top of n8n + WhatsApp AI assistants — vxdant23 · 2026-10-20
- Indie dev's anti-slop lessons for building with AI: solve your own pain, ship one thing you'll use daily — chaseleantj · 2026-10-20
- Agent sessions re-learn your project every time: devs flag recurring token waste — tekbog · 2026-10-20
- LLM masters undocumented compiler framework, raising the question: should DSLs be designed for machines? — GianniMariani · 2026-10-20
- Cursor Pro+ $60/month quota shown: Cursor Models 25% used, Other Models 21% — pramodk73 · 2026-10-20
- Cohestry: an Apple-native second brain that feeds curated, redacted context to AI agents via MCP — Emojinapp · 2026-10-20