DoorDash has shifted engineering agent workloads from developers' laptops to its Flux cloud platform, which automated 130,000 engineering tasks in a single month in 2026, according to a report published by InfoQ in August 2026. The platform handles more than 25,000 automated code reviews weekly, with over 300 playbooks and more than 10,000 weekly invocations that let workflows run without supervision and in parallel. DoorDash built Flux to overcome constraints it faced when agent workloads operated on individual laptops.
The platform operates through four core components: cloud sandboxes, an MCP gateway, reusable playbooks, and invocation surfaces, according to the report. Engineers can start agents from Slack, GitHub, scheduled jobs, the command line, or conversational interfaces. The sandboxes use Firecracker micro virtual machines to isolate agent workloads, with each environment set up with the repositories, development tools, secrets, and runtime dependencies a task needs. DoorDash reports a 95th percentile service level objective of under five seconds for complete sandbox setup, including starting the microVM, cloning repositories, installing build tools, and configuring the coding agent harness. Playbooks, written in YAML, define the task, required tools, permissions, validation, and safety limits, and can mix agent-driven steps with deterministic code where predictable execution or validation is needed.
DoorDash said local execution on laptops restricts available CPU and memory, requires the developer's device to stay connected, and can grant autonomous agents access to credentials and internal systems already available to the developer. The company also said local execution makes it harder to monitor where agents operate, which systems they reach, and on whose behalf they act. Agents reach DoorDash's internal systems through Agent Gateway, an in-house MCP gateway that gives scoped permissions and logs agent activity for auditing and policy enforcement. Radoslav Krehlik, security architect, wrote that Flux was "built after laptop-based agents ran into limits on power, uptime, and security" and that engineers can trigger agents from Slack, GitHub, or scheduled jobs while DoorDash maintains full enterprise security guardrails.
The report explains that DoorDash first used Flux for automated code review, which the company called a frequent and measurable workflow. The platform later grew to cover CI triage, on-call tasks, maintenance workflows, and ticket-driven development. DoorDash also switched its Slack integration from private channels to public threads so engineers could watch agent executions, check results, and see how other teams handed off work. The approach fits a wider shift toward cloud-based execution for coding agents—GitHub supports both local and cloud sandbox environments for Copilot, with controls over filesystem, network, and system access, and its cloud sandboxes give isolated environments for agents to run coding tasks without using local developer resources. Duy Nguyễn, co-founder of TOP GROUP Vietnam, wrote that once agents handle background work, code review, internal tools, and concurrent execution, "the hard problem shifts from model choice to control: identity, permissions, sandboxing, audit logs, quotas, and a real kill switch." As agent workloads grow in scale and scope, the infrastructure to manage them safely becomes the bottleneck, not the intelligence of the models themselves. Organizations that treat agent deployment as purely a capability question risk discovering governance gaps only after systems are already entrenched.