A harness is the machinery underneath an agent. It reads the instruction files, gives the agent its tools and connectors, runs the schedule, holds the conversation history, and delivers what the agent says to wherever it should go. Every agent runs on one.
Harnesses change fast. Several exist as of September 2026 (Hermes Agent, OpenClaw, Claude Code, GrokBot, among others) and the list will look different in a year. That is fine, because the harness is not where the value lives. The value is in the instruction files, the documentation, and the connectors, and those move.
We have moved a working fleet from one harness to another without rewriting its instructions. That is the test. If a change of harness means starting over, the agent was never yours.
Example
What it looks like.
The same publishing agent, two harnesses:Harness A: reads SOUL.md, runs the 10:15 job, delivers to a chat.Harness B: reads the same SOUL.md, runs the same job, delivers to the same chat.Nothing in the file changed.
Related
Read next.
- agentSoftware that reads its instructions, looks at the situation, takes actions with real tools, and reports back. Runs on a schedule or on a trigger, without being prompted.Read the definition
- portabilityYour agents are files. They move to any harness, any model, any host. Leaving is copying a folder.Read the definition
- modelThe intelligence an agent calls when it needs to read, write, or decide. Replaceable. When a better one ships, the same instructions run on it and the results improve.Read the definition
- connectorA link between an agent and a real system: email, calendar, files, the CRM, the site. Connecting your people to their systems is one of the first things we do in month one.Read the definition