The Orchestration Tax
TL;DR · AI Summary
Starting more AI agents is easy, but it increases the Orchestration Tax, which requires more judgment and code merging through the human processor, leading to higher costs. It is recommended to run only 1-2 tasks at a time, focus, and optimize the process.
Key Takeaways
- Starting more AI agents increases the Orchestration Tax
- Human processors have single-threaded attention and cannot parallelize
- It is recommended to run only 1-2 tasks at a time, focus, and optimize the proce
Outline
Jump quickly between sections.
The article discusses the ease of starting more AI agents and the resulting Orchestration Tax.
Human processors have single-threaded attention and cannot parallelize, requiring all AI agent judgment and code merging to go through the human processor.
Starting more AI agents increases the Orchestration Tax, leading to higher costs.
It is recommended to run only 1-2 tasks at a time, focus, and optimize the process to improve productivity and control.
Mindmap
See how the topics connect at a glance.
查看大纲文本（无障碍 / 无 JS 友好）
- 启动AI代理的编排税
- 人类处理器的限制
- 单线程注意力
- 编排税
- 成本升高
- 解决方案
- 同时运行1-2个任务
Highlights
Key sentences worth saving and sharing.
Starting more AI agents increases the Orchestration Tax, which requires more judgment and code merging through the human processor, leading to higher costs.
Viking on X: "Great article! Starting AI agents is easy, but human attention is single-threaded and can't be parallelized. It has limited context handling and context switching is difficult. All the agents' final decision/merge code has to go through your weak processor. So, the more agents you start, the higher the cost you pay later. He calls it Orchestration Tax, the tax of orchestration. So, I now only run 1-2 tasks at a time."
Don’t miss what’s happening
Show translation
Great article! Starting AI agents is easy, but human attention is single-threaded and can't be parallelized. It has limited context handling and context switching is difficult. All the agents' final decision/merge code has to go through your weak processor. So, the more agents you start, the higher the cost you pay later. He calls it Orchestration Tax, the tax of orchestration. So, I now only run 1-2 tasks at a time, carefully review AI-generated code, optimize processes, and focus on one task at a time. The article has a great point: categorize tasks into simple asynchronous ones versus those requiring deep judgment, and don't force complex tasks to be parallelized. I've tried working on up to 4 different projects simultaneously, which felt very tiring, with no productivity increase and no sense of accomplishment or pleasure. In the long run, I also felt a lack of control over my projects. Of course, mindless vibe coding doesn't count.
Quote
Addy Osmani
@addyosmani
·
May 28
Article
The Orchestration Tax
Starting more AI agents is easy now. However, more agents running doesn’t mean there's more of you available - your cognitive bandwidth doesn’t parallelize. All the judgement to actually steer them...
·
108
11
56
63
Read 108 replies