The AI Explainer, Part 3: Agents and Harnesses
The model is not the product. The product is the software system built around the model. Part 3 looks at the wrapping that does most of the work everyone calls 'AI.'
The AI Explainer, Part 3: Agents and Harnesses
The model is not the product. The product is the software system built around the model. A chat box is a model plus a thin prompt. A coding assistant is a model plus a tool loop plus a state machine plus a sandbox plus a reviewer. The model itself is a frozen function. Everything else lives in code wrapped around that function.
Part 3 of the trilogy looks at the wrapping. What an agent actually is. The tool loop and who drives it. Why most "AI" in production is a workflow, not an agent. Why "harness" won the naming fight for the surrounding software. And evaluation as the only honest way to know if any of this works. Five articles on the layer that does most of the work everyone calls "AI."
In This Series
- 1
An Agent Is Not a Smarter Model
The model is a frozen function. Most of what people call 'AI' is the software wrapped around it.
4 min read - 2
The Tool Loop and Who Drives It
Tool use is what lets a frozen language model take an action in the real world. The interesting question is not how it works, but who controls the loop.
4 min read - 3
Most 'AI' in Production Is a Workflow
Anthropic's own engineers tell you not to build an agent when a workflow will do. The deployed wins prove them right.
5 min read - 4
Why 'Harness' Won the Naming Fight
By mid-2026, the software wrapping a model has a name. Most of what you call 'AI' is the harness doing the actual work.
5 min read - 5
A Demo Proves Nothing
Frontier AI agents fail more than half the time on real customer-service tasks, and the variance is worse than the average. Evaluation is the only honest answer to 'does it work.'
5 min read