🏡

👨‍🎨

🤖

📚

📝

 

'Harness'

Published: Sep 13, 2026

The word harness, in the context of language models, has specific and opinionated connotations, namely that in its raw and unharnessed state the language model is a horse, unbridled, free, and it needs to be tamed and subjugated in order to be made useful.

This seems broadly reasonable, although I wonder if by some accident of history we could have gotten a different name. Below are some possible candidates, along with their respective vibes.

• Trellis or scaffold -- the model grows organically around a pre-defined structure, and it grows upwards, towards light or towards completion of the final building. These give a better sense for the unpredictability of the endeavor of using LLMs for work, with sunlight and the blueprint as a metaphor for user intent.
• Actuator or body -- the raw model is a brain in a vat, with no means of influencing the world. It needs a body in order to actually actuate change and do things in the world.
• Exoskeleton -- harness implies that the model is a horse, and exoskeleton implies that it is either human or insect, both of which have greater range of motion --can climb, can lift heavy things with their limbs-- than horses.
• Generator -- the model is a source of energy, and that energy is turned into intelligence and work for the world, via the generator. A bit more confusing, since GPTs already have generate in their name.
• Kitchen -- the model is a chef, and is given ingredients and tools and asked to prep a meal for consumption.

If you think of more, please email me!