DevDay 2026 read: building self-verifying agents with Sol and the Agents API
The article argues DevDay's real story is not a model release but GPT-6.1 Sol plus the Agents API, computer use and subagents, which can build agents that persist, act and check their own work.
What this article says
作者认为多数人把 DevDay 2026 当成又一次模型发布，但真正的变化在于代理形态：GPT-6.1 Sol 提供推理内核，Agents API 提供编排层，computer use 让它能直接操作界面，subagents 则把长任务拆成可并行的子任务。四者叠加后，代理不再是一次问答，而是能持续存在、主动执行、并对自己的产出做验证的循环。文章随后给出分步搭建方法，从定义任务边界、配置工具与权限，到设置子代理分工和自检环节，目标是让代理在无人盯着的情况下也能交付可核验的结果。结论是：这一轮竞争的分水岭不在跑分，而在谁能把「持久、行动、自验证」这三件事工程化。
Most people will see DevDay 2026 as another model release
They're missing the real story
GPT-6.1 Sol + Agents API + computer use + subagents = an agent that persists, acts and verifies its own work
Here's how to build one, step by step https://x.com/i/article/2101672607179284480