AI 采用的下两个层级：从编排走向模拟
The Next Two Levels of AI Adoption
Every 在 6 月提出的 AI 采用八级框架之后补充了第 9、10 级：第 9 级让不同策略相互竞争以找出最优解，第 10 级反复运行这一竞争以识别能稳定胜出的策略。作者称多数工程师已在同时运行多个长任务智能体并把协调交给编排器，Codex 和 Claude 桌面应用让非技术团队也能更快推进编排。这两级仍处于实验阶段，消耗的 token 远多于低层级，且更高层级并不保证当下就有更好结果。
What comes after orchestration?
Nov 12, 2026 · 9 min read
Was this newsletter forwarded to you? Sign up to get it in your inbox.
In June, we published the eight levels of AI adoption, giving people a map of where they are with AI today and what it takes to get to the next level. The map ran from Level 1, where you ask a chatbot a question, to Level 8, where a manager agent runs a team of subagents for you. I expected most readers to find themselves in the bottom half, and most of them did. What I didn’t expect was how quickly the people at the top would reach the end of the map.
Today, most engineers I talk to are running several long-running agents at once and handing the coordination work to an orchestrator. Even nontechnical teams are progressing faster than I thought they would, because the Codex and Claude desktop apps make orchestration much easier for knowledge work. As more people approach the top end of the eight levels we outlined in June, I keep getting asked, “What comes next?” My answer, which I covered in my talk at Thesis: 2027 last week, is simulation: building a model of the world your agents work in, so you can test strategies before reality does.
In the original framework, each level let you hand off longer stretches of work and gave the AI more autonomy, with orchestration representing the most you could hand off. Today, there’s no reason to believe that pattern should end with orchestration. Once you have agents managing and executing the work, the harder question becomes which strategy to bet on. Levels 1–8 increased delegation and autonomy. Levels 9 and 10 extend how agents explore and test strategies. Level 9 pits strategies against each other to find the best one, and Level 10 runs that competition multiple times to see which strategies win consistently.
Here they are, with some caveats: These are firmly in experimental territory and burn many more tokens than lower levels, and as always, working at a higher level doesn’t guarantee better results today—it’s an investment in tomorrow.
来源：Every：最新文章 · every.to