Business Pill 15 · The model and the harness
Two products built on the same AI model can perform very differently. The difference is the system around the model, which has a name: the harness.
A short explainer video. The numbers in it are round numbers for illustration.
What a Model Does
A model does one thing: it takes text in and it predicts text out. On its own it cannot open a file, it cannot remember yesterday, and it cannot check its own work.
What the Harness Is
Everything else is built around the model: the instructions it is given, the tools it can use, the memory it keeps between steps, the loop that lets it try, look at the result and try again, and the checks that catch its mistakes.
That surrounding system has a name. It is called the harness, and it is what turns a model into a worker. Think of an engine and a car. The engine provides the power, but steering, brakes and a dashboard decide whether you actually arrive. Two cars with the same engine can drive very differently.
A Worked Example
The video offers an illustration. Give the same model 10 tasks with a bare set-up and it finishes four. With a careful harness, the same model finishes eight. Nothing about the model changed.
It also works the other way. Put a stronger model inside a weak harness and most of the gain is lost. A model can only be as useful as the system around it allows.

Why It Matters
The video gives two reasons. First, models can be swapped. When a better one arrives, a good harness simply plugs it in. Second, the harness is where a company’s own knowledge lives.
Three Questions to Ask
- Which model is inside?
- What has been built around it?
- Which of the two is harder to replace?
See It in the News
Google Research Cogentic Reports Five Open Math Results. A news piece about a research harness built around a model: an orchestrator, provers and adversarial verifiers.
Claude Opus 5 Scores 97.5% and 30.16% Simultaneously. A news piece about how a benchmark result depends on the scaffolding around the model.
More Business Pills
- The Memory Wall: Why a Faster AI Chip Is Not Faster AI
- Tokens per Watt: What an AI Data Centre Actually Produces
- Why AI Can’t Be Both Instant and Cheap: Latency vs Throughput
- Context, Not Capability: Why a Smart AI Model Gives Poor Answers
- Discardable Software: When Code Is Cheap Enough to Throw Away
- Human in the Loop vs Human on the Loop: Supervising AI Agents
- The AI Audit Problem: When Making Work Is Cheaper Than Checking It
- AI Evaluation as Acceptance Test: How to Know It’s Good Enough
- Extensibility Is Control: Who Holds the Power in an AI Product









