AI Agent Teams Can't Agree on Basic Terms Yet

May 27, 2026

AI Agent Teams Can't Agree on Basic Terms Yet

Published: May 27, 2026 at 12:09 AM

Updated: May 27, 2026 at 12:09 AM

100-word summary

Hugging Face published a glossary in May 2026 trying to settle what "agent," "harness," and "scaffolding" actually mean. Their breakdown: the model is the LLM itself, scaffolding shapes behavior around it, and the harness is the execution loop running tool calls. An agent is the full system combining model plus harness. The problem? These definitions still vary wildly across frameworks. There's no shared vocabulary yet, which means teams inside the same company often talk past each other. Hugging Face argues separating harness from scaffolding matters because one drives runtime and the other shapes context. The bigger reveal: the AI industry is moving fast enough that we're building complex systems before...

What happened

Hugging Face published a glossary in May 2026 trying to settle what "agent," "harness," and "scaffolding" actually mean. Their breakdown: the model is the LLM itself, scaffolding shapes behavior around it, and the harness is the execution loop running tool calls. An agent is the full system combining model plus harness. The problem? These definitions still vary wildly across frameworks. There's no shared vocabulary yet, which means teams inside the same company often talk past each other. Hugging Face argues separating harness from scaffolding matters because one drives runtime and the other shapes context.

Why it matters

The bigger reveal: the AI industry is moving fast enough that we're building complex systems before agreeing what to call the pieces.

Sources