TechnicalAI glossary
Agent harness
The software wrapped around a model to turn it into an agent: it gives it tools (reading files, running code, searching), runs the observe, decide, act loop, manages what it keeps in memory and sets what it is allowed to do. The same model gets different results depending on the harness, which is why leaderboards now compare model + harness pairs.
In other words
The same engine fitted in a racing car or a delivery van: the horsepower is identical, the lap times far less so.
Alyriaz articles that mention it
- The model rankings as of 23 September 2026
State of the model landscape as of 23 September 2026: who leads on what, at what price, and how to read leaderboards that have become hard to compare.
- How an agent navigates the web
A browsing agent chains together searches, reads and cross-checks to find hard-to-find information. We explain the loop, where it fails, and how to use it properly.
- A thousand engineers deployed at customer sites to install agents
Accenture and Google Cloud are creating a dedicated entity for deploying enterprise agents, with a thousand engineers deployed at client sites. We analyse what this model reveals about the real obstacle to adoption.
- A model that writes a text in one go instead of word by word
Diffusion models applied to text generate an entire response in parallel rather than word by word. We explain the principle, its speed advantages and its limitations.
- K2 Horizon: six open models, from 0.9 to 375 billion parameters
The Institute of Foundation Models has released six open models under Apache 2.0, spanning a range from 0.9 to 375 billion parameters. We explain why publishing a full range changes the game for those building.
- What is an agent "harness"? The wrapper that does half the work
Agent evaluations now measure a model + harness pairing. We explain what this software wrapper is, why it weighs as heavily as the model, and what it changes when choosing.