DoorDash Runs 130K Agent Tasks in Modal
DoorDash reportedly ran 130,000 agent tasks in a month using locked-down Modal Sandboxes, giving its production agents isolated execution environments without granting direct access to core systems. The deployment shows secure runtime infrastructure becoming a prerequisite for serious agent adoption.
DoorDash’s usage is a stronger signal than another sandbox launch announcement: agents are already operating at meaningful enterprise volume when the execution boundary is designed correctly.
- –Isolated sandboxes let agents run code and tools while limiting filesystem, network, and credential exposure.
- –Modal provides the burstable infrastructure needed for high-volume, asynchronous agent workloads.
- –The real unlock is operational confidence: teams can let agents act without treating every task like a production deployment.
- –Developers building agents should prioritize runtime policy, observability, rollback, and scoped access alongside model quality.
- –DoorDash’s example suggests the debate is shifting from whether agents should execute work to how safely that execution can scale.
DISCOVERED
2h ago
2026-08-12
PUBLISHED
4h ago
2026-08-12
RELEVANCE
AUTHOR
jecrosbie