Two of the people who built Kubernetes think AI agents need the same treatment containers got in 2014. Stacklok, the company founded by Kubernetes co-creators Craig McLuckie and Joe Beda, has open-sourced Mecatl, a cloud-native agent harness designed to run AI coding and task agents as production workloads on infrastructure the enterprise operates — not on the developer\'s laptop.
The core architectural bet is separation. Mecatl keeps the agent loop independent of the client, the model provider, the state store and the execution environment, so the same runtime can start locally, move to a remote server with durable external state and an append-only event log, then scale across Kubernetes replicas without rewriting the loop. The reference runtime pairs Redis for session state with Kubernetes leases to guarantee one writer per session, plus a drain path for replacing pods mid-task.
Security gets first-class treatment rather than bolt-on scripts. Mecatl ships deny-dominant permissions, approval flows, secret-scrubbed command environments, durable attribution and an audit trail; delegated runs receive derived capabilities that can only narrow at each hop. Execution environments are per-session, including an opt-in local microVM backend that keeps model credentials on the host while filesystem tools and Bash run inside the VM. The project also includes mecak8s for Kubernetes deployment, mecatui as a terminal client, a TypeScript SDK, and gRPC plus HTTP/SSE integration interfaces.
The founders\' argument is about where enterprise value lives. "That\'s where the IP sits," Beda said of developer machines, where code and context originate and where agents are hardest to monitor, isolate and govern. His objection to the obvious alternative — lifting a desktop harness into a VM or container — is that it breaks agent lifecycle management, starting with quiescing: safely pausing an agent while it waits for a human. Beda likened enterprise agent interactions to email, an activity organisations expect to manage, retain and audit as a system, not as thousands of individual desktops.
Stacklok\'s route here runs through its earlier products. The company pivoted from software supply chain security to agentic infrastructure and built ToolHive, an open-source platform for running and governing MCP servers that grew from desktop Docker management into a Kubernetes gateway, registry and operator. Its AI Gateway — access control, budgets, reporting and provider routing, though deliberately not model selection — is the one major product not yet open-sourced. McLuckie says Stacklok\'s commercial layer is "the enterprise spine, the control plane" tying the open-source pieces together with consistent identity, authorization, policy and auditing, and that its customers are concentrated in regulated industries: banks, semiconductor companies and telcos.
There is a strategic subtext, too. McLuckie frames Mecatl the way Kubernetes is usually framed — as an escape hatch from platform lock-in, in this case from frontier labs and hyperscalers whose bundled harnesses arrive with their own gravity. Whether regulated enterprises actually move agent workloads onto Kubernetes at scale is the open question; Stacklok raised a $17.5 million Series A in 2023 from Accel, Madrona and Bain Capital, and is betting the answer arrives within the current agent buildout cycle.
The timing is not accidental. OpenAI and Anthropic have both spent 2026 pushing their own harnesses into the cloud with mixed results, and the space between "a terminal app on a laptop" and "a governed production workload" is exactly where most enterprise agent deployments currently stall. Mecatl is an attempt to make that middle ground somebody\'s product instead of everybody\'s problem.
Comments (0)
Log in to join the discussion
Log InNo comments yet