Intel / video

When to Build Your Own Agent Harness | Harrison Chase, LangChain

Source: youtube.com
Author: Sequoia Capital
Date: Published Sep 3, 2026

Context

Harrison Chase, co-founder and CEO of LangChain, describes an as the code around a model that decides what it sees and what happens with its response, in a loop that calls tools. Ready-made harnesses such as Claude Code or Codex work with particular vendors' models. In a Sequoia talk, Chase offers a rule of thumb, not a measured result: the more your task resembles what the models were trained on, the better an off-the-shelf harness will serve, and the further away it is, the more you should tune your own.

The example is legal AI. The overall task is far from model training, but editing files is not, and Chase says OpenAI and Claude models were each trained to edit files in different ways. The suggestion is a custom harness that still uses each model's own edit tool. Chase says LangChain's Deep Agents does this with model profiles that switch the edit implementation by model.

Terms in this piece · Glossary

agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.

Read the source www.youtube.com