Why two AI assistants on the same model act so differently
Nielsen Norman Group, a long-running user experience research firm, published a definition of "agent harness": the software wrapped around an AI model that runs its work loop and decides what it can see, which tools it can use, and where its limits are.
The same model can feel capable in one app and useless in another, and the harness is usually the reason. If you're getting into tech, this is a term you'll keep hearing. If you run your own business, it means judging an assistant by whether it finishes a real job, like drafting and sending a follow-up, rather than by how polished its answers sound.
Worth a few minutes if AI tools confuse you or you're choosing one to pay for. Skip it if you already compare assistants by running real tasks through them.
Your moveRead the short definition, then give two AI assistants the same real task from start to finish and note which one actually completes it.
Read the source