Harness / Model Separation
The agent harness and the model that drives it are separable artifacts. The harness is everything that is not weights: the system prompt, the tool definitions, the agent loop, the permission model, the file conventions, the subagent machinery. The model is an interchangeable inference backend behind them. The source's analogy — Claude Code is the car; the chat model is the engine — is doing real work, because the two are usually bought together and therefore usually reasoned about together. Pulling them apart reframes what you are paying for: when an operator hits a "context limit" or a "token limit," that is a property of the model contract they bought, not of the harness they are using. The harness itself is a piece of software running on their machine, and it will point at whatever inference endpoint they tell it to (see BYO Inference Endpoint).
This separation is what makes Model-Tier Routing mechanically possible at all — you cannot route a sub-role to a cheaper tier unless the loop around that sub-role is indifferent to which model answers. It is also why the harness, not the model, is increasingly where vendors differentiate: the loop is reproducible, but the accumulated tool design (Agent-Computer Interface (ACI)), prompt scaffolding, and conventions are the durable artifact.
Separability is not free, and the failure mode has its own page: a harness encodes an implicit contract that the substituted model must satisfy — see Harness / Model Fit. The engine bolts in; the car may still drive badly.
Claims
- An agent harness and the model that drives it are separable artifacts; the harness supplies the tools, the loop, and the instructions, and the model is swappable underneath it. principle — durable: it follows from the harness being ordinary software that speaks an inference API, independent of vendor, model generation, or which harness. The bundling is commercial, not architectural.
- Usage limits, context limits, and token bills are properties of the model contract, not of the harness. principle — durable: they attach to whoever serves the weights. The operator who conflates the two will believe the harness is the thing costing money.
- The harness — tools, loop, conventions, scaffolding — is the accumulated, differentiating artifact; the model behind it is the commodity input. observation — the direction the source's demo implies (he keeps Claude Code and discards the Anthropic engine) rather than a claim he argues. Related to Execution Commoditization, which makes the general form of this argument.
- The source states that pointing a vendor's harness at a third-party model is not a terms-of-service violation, reasoning "we're using their agent harness, we're just plugging in a different model." observation — asserted on a title card without citation; a licensing question, groundable and explicitly not adjudicated here. An operator relying on this should check it directly.
Related
- Harness / Model Fit — the necessary counterweight: separable does not mean interchangeable. The harness carries an implicit contract (tools, context window, protocol) that a swapped-in model can silently fail.
- BYO Inference Endpoint — the concrete config seam through which the separation is exercised.
- Agent Loop — the part of the harness that survives the swap unchanged.
- Agent-Computer Interface (ACI) — the tool definitions are the harness's most valuable half, and the half that assumes a model trained to use them.
- Model-Tier Routing — routing sub-roles across model tiers presupposes this separation; a self-hostable bottom tier is only reachable because the loop is model-indifferent.
- Open-Weight Capability Gap — what determines whether a given engine is worth bolting in.
- Execution Commoditization — if the model is the commodity, the harness and the method around it are where value lands.
- Distillate: Ollama + Claude Code = 99% CHEAPER