# Cognition发布SWE-check并阐述AI工程新范式

- 来源：swyx 🐣 (@swyx)
- 发布时间：2026-04-15 10:01
- AIHOT 链接：https://aihot.virxact.com/items/cmnzgrdaj03g0sl0fhocofre3
- 原文链接：https://x.com/swyx/status/2044234661644382581

## AI 摘要

Cognition发布基于强化学习的bug检测模型SWE-check，在匹配前沿模型性能的同时实现10倍推理加速。作者提出AI工程的核心范式：通过模型与工具组合推动AI帕累托前沿，而非直接突破模型边界；应采用"先最大化能力再蒸馏"的策略。Applied Compute正为多家Agent Lab提供算力基础设施。AI领域仅存在两种商业模式：能力整合与能力拆分。

## 正文

proud to see @excalidraw evangelism catching on at cog

the insight here is more general than bugchecking：
- All Engineering is about making tradeoffs
- AI Engineering is about pushing AI Pareto Frontiers with any combo of model + harness at your disposal
- Don't try to directly break a model frontier - instead you should first capabilitymaxx， then distil
- this works ~basically every time 【citation needed】
- @appliedcompute is arms dealer to every Agent Lab doing this sort of thing rn， it's really fascinating to see this deployed on every high volume AI problem

only 2 ways to make money in AI： bundling capabilities， and unbundling them！

### 引用推文

> Cognition：Today we're releasing SWE-check, a specialized bug detection model we RL-trained with @appliedcompute that matches frontier performance on internal in-distribut...