DoorDash's Flux automates 130,000 engineering tasks in a month
DoorDash has moved engineering agent workloads off developer laptops onto its in-house cloud platform Flux, which automated 130,000 engineering tasks in a single month in 2026 and handles more than 25,000 automated code reviews a week, with over 300 Playbooks and more than 10,000 weekly invocations, according to InfoQ Chinese. Flux rests on four platform primitives: cloud Sandboxes isolated with Firecracker micro virtual machines, an internal MCP gateway called Agent Gateway that scopes and logs access to internal systems, reusable YAML-defined Playbooks, and invocation surfaces spanning Slack, GitHub, Cron and the command line; DoorDash says its 95th-percentile target for end-to-end Sandbox setup is under five seconds. DoorDash started with automated code review, then extended Flux to CI triage, on-call, routine maintenance and ticket-driven development, and moved its Slack integration from private channels to public threads so engineers can watch agent runs.
A commenter quoted in the report argues that once agents handle background work, code review and concurrent execution, the hard part shifts from model choice to control: identity, permissions, sandboxes, audit logs, quotas and a real kill switch.
DoorDash已经将工程智能体工作负载从开发者的笔记本迁移到其Flux云平台"上，该平台在2026年单月自动化完成了130000个工程任务。该平台每周支持超过25000次的自动化代码审查，拥有超过300个Playbook和每周超过10000次的触发调用，允许工作流无人值守并行运行。
DoorDash在其智能体工作负载在单台笔记本上运行时遇到限制后开发了Flux。本地执行受限于可用的CPU与内存，依赖开发者设备保持在线，并可能让自治智能体访问开发者已拥有的凭据与内部系统。DoorDash还表示，本地执行使得监控智能体的运行位置、它们访问的系统以及它们代表谁操作变得更难。
#DoorDash#Flux#Firecracker#Agent Gateway#Model Context Protocol#GitHub Copilot#Slack