Blog

AI Glossary Update, October 2026: From Prompt Engineering to Harness Engineering

The words for building with AI keep changing. This update covers harness engineering, why prompt engineering is now one part of a bigger job, and what to call AI search visibility.

By - 2026-11-17

AI Glossary Update, October 2026: From Prompt Engineering to Harness Engineering

The words for building with AI change fast. The KPT Social glossary started in February 2026 with 346 terms and added 30 more in June. This is the first of a regular series of updates: what changed in the field, what changed in the glossary, and what it means for a small business. Old updates stay up, so you can see how the field moved.

What changed in the field

Harness engineering got a name

For years, people building AI agents did the same work without a shared word for it. In February 2026 it got one: harness engineering. OpenAI published an engineering report on it the same month, and the shorthand that stuck is "Agent = Model + Harness" (awesome-agent-harness).

The harness is everything in an AI tool except the model: the loop that keeps a task moving, the context the model sees, the tools it can call, its memory, its permissions, and the checks it runs before work reaches a person (Perplexity). Harness engineering is designing that layer so the model does useful, reliable work (Snowflake). The idea has moved into vendor documentation (Microsoft Learn) and into policy research, where a United Nations University report treats the harness as the layer to govern (UNU).

Prompt engineering became one part of the job

Prompt engineering is not gone. It's now the inner ring. One common way to put it: prompt engineering shapes a single turn, context engineering manages what the model sees, and harness engineering governs the whole environment the agent runs in, across every session (awesome-agent-harness). Newer models need less hand-tuned wording. What they need is the right tools, the right information at the right moment, and clear limits.

AI search still has no settled name

Getting found and recommended in AI answers goes by several names: GEO (generative engine optimization), AEO (answer engine optimization), LLMO, or "AI SEO". There is no single agreed term. GEO is the one large firms use most (Evergreen Media). Google's view is that optimizing for its AI search features is still SEO (Google Search Central). We use GEO in the glossary and "AI search visibility" in plain language.

What changed in the glossary

Added

  • Harness Engineering: the discipline of building everything around an AI model so it can do reliable work.

Updated

Kept, with a note

What this means for a small business

If you're buying AI help, ask about the harness, not only the model or the prompts. Questions worth asking:

  • Which tools can the agent use, and which can't it touch?
  • What information does it see for each task, and where does that come from?
  • What does it check before anything reaches a customer?
  • When does it stop and hand off to a person? (See human-in-the-loop.)

Those answers decide whether an AI system is dependable. For a deeper walk-through, read AI Harness Engineering: Complete Guide for Small Businesses and AI Harness Engineering: Small Business Automation Guide.

Every build starts with the AOF Blueprint, which maps these decisions for your business before anything is built.