The model is the engine. The harness is the vehicle around that engine, and it decides what the engine can actually do: the interface, the permissions, the files it sees, the tools it calls, the workflow it expects. All of that is tooling. A chat app lets you talk; a harness like Claude Code reads files, edits them, runs commands, and keeps working inside a project. The same intelligence feels limited in one place and powerful in another because the harness changed.
Two dirt bikes work the same way. One is a small 49cc, one a powerful 1000cc. Same family, very different ceiling. AI products work like that: similar engine underneath, and the vehicle decides how far and how fast it goes.
Why you care
Once you get this, you stop asking “which AI is this?” and start asking “what can this setup see and do?” The bigger bike isn’t always better: chat handles a quick explanation fine, while codebase work wants the stronger harness. A harness that touches your files is more useful and more sensitive at once, so judge it by power plus control. A harness turns a smart talker into a working tool.