Загрузка видео...
Не удалось загрузить видео
Agent harnesses explained in <90 seconds. TLDR: if you’re not working on the model, you’re working on the harness.
15,584 просмотров • 11 дней назад •via X (Twitter)
Комментарии: 8

模型还没搞明白,套件先学会了也挺真实的

What's going ond Sounds like someone's trying to describe a new way of understanding something!.

@grok comment faire une vidéo comme ça

The harness distinction is key - reliability, permissions, and tool loops are where agent products become usable.

Yes. We spent months in the harness and the thing that kept biting us was the agent opening pages it had no business opening. No harness had a map of the web, so we built one. AI agent allow list: 40M domains, 28 page types, login, checkout and upload denied before the tool call.

这句很到位,模型会换,但上下文、工具和失败处理才是最磨人的那层。

this harness vs model split is so real right now. most teams i see shipping agents are just fighting the orchestration, memory, tool calling and eval loop anyway. the model is almost becoming a commodity at this point. good explainer tho, gonna check that academy course later.....

A harness becomes production-grade when each run emits a replayable trace: model and tool versions, arguments, state transitions, latency, cost, and evaluator verdict. Keep a small failure corpus in CI so harness changes are tested against regressions, not just happy-path demos.
