Vibeleaderboard
← All Intel
Intel / video

Google DeepMind's Logan Kilpatrick: Why the Model Eats the Harness

Source
youtube.com
Author
Sequoia Capital
Date

Context

Logan Kilpatrick, who runs Google AI Studio and the Gemini at Google DeepMind, argues in a Sequoia Capital interview that the software wrapped around a model tends to get absorbed by the model over time. Kilpatrick describes the product called 'the model' as no longer just weights but an expanding system including hosted tools, containers and an , the that lets a model take actions.

The pattern Kilpatrick describes is that scaffolding often runs a few steps ahead of what the model does natively, then the model absorbs it. Within about 12 months, Kilpatrick expects, building your own will probably no longer be where the advantage lies, though external search and code execution may stay useful. On whether companies build harnesses to avoid lock-in, Kilpatrick says that may start true but fades as models improve, and suggests a 'harness bench' measuring how well models adapt to different harnesses.

Kilpatrick also says the Antigravity agent harness will power agent features across Google products, including search and the Gemini app. All of this is a Google DeepMind product lead's view and a prediction, not a result.

Terms in this piece · Glossary
  • API — Application programming interface: the agreed set of operations, inputs and outputs through which one piece of software uses another.
  • Harness engineering — Designing the full environment and feedback system around an agent.
  • Program — A set of instructions stored so that a computer can run it.
  • Harness — The software that sends a model its context, recognizes tool calls, runs or refuses them, returns results and keeps the record.
Read the source www.youtube.com
More from Sequoia Capital
Recommended reads
Comments

Checking sign-in…

Loading comments…