Can a Cloud-Native Harness Make Agents Reliable Beyond the Desktop?
- Source
- latent.space
- Author
- Richard MacManus
- Date

Context
Stacklok, a startup led by Kubernetes creators Craig McLuckie and Joe Beda, is building Mecatl, an open source coding- meant to run in the cloud instead of on a laptop. A harness is the program that loops a language model through tool calls while carrying forward. In Latent Space's report, Beda says desktop-first harnesses typically start with the loop, local execution and session state in one process, and that moving them into a container or virtual machine causes lifecycle problems, such as not being able to safely pause an agent that is waiting for a person.
Mecatl keeps the agent loop independent of the client, model provider, state store and execution environment, so commands, session management and memory can live in separately managed systems rather than files on disk. Stacklok targets large enterprises, which Beda expects to govern agent interactions the way they govern email. Its commercial product is a control plane for identity, policy and auditing. These are the founders' claims about a project begun in June; the article reports no comparison with other harnesses.
- Mecatl, Stacklok's open-source harness started in June, keeps the agent loop independent of client, model provider, state store and execution environment, and splits out tool calling, bash, sessions and memory instead of keeping JSONL files on disk.
- Joe Beda says lifting a desktop harness into VMs, containers or sandboxes causes lifecycle problems, such as safely pausing (quiescing) an agent while it waits for human input.
- ToolHive, Stacklok's open-source platform, began as a way to run MCP servers in Docker on the desktop and grew into a Kubernetes-based gateway, registry and operator.
- Its AI Gateway, not yet open source, handles access control, budgets, reporting and provider routing but not task-based model choice; Craig McLuckie says that routing belongs in the harness, which has more context.
- Stacklok sells the "enterprise spine": identity, authorization, policy and auditing across the open-source parts. McLuckie says most customers are banks, semiconductor companies and telcos.
- Harness engineering — Designing the full environment and feedback system around an agent.
- Context and context window — Context is the information available for a generation; the context window is the capacity available to hold it.
- Terminal and shell — A text interface for navigating files and running programs.
- Model Context Protocol (MCP) — A published agreement for an AI application to discover what a separate program offers and ask it to run one of those offerings.
articleThe War Against Slop and an Emerging Weapon to Fight it: Formal Verification
articleCloudflare Joins the Jevolution and Labels Itself a Neolab
articleThe Jev Revolution: What Engineers Are Already Building with System One Models
articleAI Engineer Code Summit: OpenAI on New Ways of Working With Codex
articleHarness Engineering: Anatomy, Architecture, and Evolution of Coding Agents -- A Source-Code Study of Eleven SystemsPaul Barbaste, Tristan Darrigol, Germain Vu, Tom Wiltberger
videoHarness Engineering: Building the Production Cage for Powerful Domain Agents — Mike Chambers, AWSAI Engineer
articleMaking every harness durablePranay Prakash
Checking sign-in…
Loading comments…