DoorDash has shifted its engineering agent workloads from developer laptops to its internal cloud platform, Flux, achieving the automation of 130,000 engineering tasks in a single month. The company also reported that the platform now supports more than 25,000 automated code reviews each week.
The move, detailed in a company announcement, centralizes agent operations within a managed environment, rather than relying on individual developer machines. This change allows DoorDash to run agent workflows with scoped access, centralized auditing, and higher scalability.
How Flux Works
Flux operates using isolated Firecracker microVMs, which are lightweight virtual machines designed for secure and efficient execution of code. The platform also includes an MCP gateway that manages communication between agents and internal tools, and reusable playbooks that standardize common engineering workflows.
Agents can be invoked through multiple surfaces, meaning they can be triggered from different points within the development process. According to DoorDash, this architecture provides control over what each agent can access and ensures that all actions are logged for audit purposes.
Implications for Engineering Productivity
By moving agents to the cloud, DoorDash aims to improve consistency and reliability of automated tasks, including code review, testing, and issue triage. The shift reduces dependency on local computing resources and enables agents to run at scale without performance interruptions.
The company did not disclose the exact number of engineering staff involved or the total time saved, but emphasized that the platform has become a core part of its development workflow. The automated code reviews alone represent a significant portion of the company’s review activity, though DoorDash did not specify what percentage of total reviews are now handled by agents.
Broader Context
The announcement reflects a broader industry trend toward centralized, cloud-based AI and automation infrastructure. Many organizations are moving away from running agent models on local machines to managed platforms that offer better security, compliance, and resource management.
DoorDash’s approach is notable for its use of Firecracker microVMs, a technology that was open-sourced by Amazon Web Services and is commonly used in serverless computing. The choice suggests a focus on high isolation and low overhead for agent execution.
Analysts say that such moves could set a precedent for other large technology companies that are exploring similar automation strategies, particularly in areas like code review and maintenance tasks.
DoorDash has not announced a public release date for Flux, as it is currently an internal platform. However, the company said it will continue to expand the number of tasks handled by agents and improve the system’s capabilities in response to developer feedback.
