Summary of Key Points
This article documents the author's shift in perspective on how AI can help organizations improve efficiency: from doubting the business viability of using AI for efficiency improvements in 2024 (believing that such efforts were not a good investment), to attempting to automate tasks with AI in 2025, to realizing that AI cannot solve “open-ended problems” (which require human creativity and subconscious ideas). The author ultimately proposes two types of AI-Native organizations—“AI Control Centers” for handling open-ended issues (where humans drive AI) and “Agent Platforms” for handling closed-ended tasks (where AI drives humans), emphasizing the irreplaceability of humans in creative work.
1. From “AI Efficiency is Not a Good Business” to “Reorganizing Work”: The Author's Cognitive Shift
In 2024, although the author increased coding efficiency by 30% using tools like Copilot/Cursor, he realized that a 30% improvement was not the same as a several-fold increase, given the complexity of real-world tasks. More importantly, he argued that the business model behind using AI for efficiency improvements had flaws: the marginal cost of using AI did not decrease (for example, adding more tasks did not reduce costs), and the benefits from increased efficiency were limited (once efficiency reached a certain level, further improvements were minimal), making it a less profitable venture.
It was not until 2025, when he encountered Claude Code, that the author discovered AI’s ability to perform tasks previously impossible. This led him to explore the correct approach to using AI for organizational efficiency: not just providing tools to enhance existing efficiency but breaking down work into tasks that AI could handle independently (more than 90%) and then gradually automating those tasks with Agents. At this point, he even wondered if companies would still need humans in the future, but he soon accepted the inevitable trend.
2. Jack Dorsey’s “AI-Centric” Organizational Philosophy: Turning a Company into an Intelligent Entity
Among the AI-Native organizations discussed in Silicon Valley, Block CEO Jack Dorsey’s approach is the most radical: transforming the company itself into an intelligent entity with three key principles:
1. Digitize all work outputs: Everything related to work must be recorded in text, code, plans, etc., so that AI can process it effectively (providing sufficient context).
2. Quick and strong customer feedback: Customer usage, payments, complaints, and other data should be promptly fed into AI to allow immediate evaluation of decisions.
3. Humans as peripheral tools: Humans are no longer at the center of decision-making but act as “real-world interaction tools” for AI; for example, if AI needs authorization to execute a plan, it requests human approval, and if the plan fails, AI automatically assigns tasks related to fixing the issues.
The author initially agreed with this philosophy, as it aligned with his own attempts to use Agents to automate tasks. However, he later realized the limitations of AI.
3. Open-Ended vs. Closed-Ended Problems: The Limits of AI
In practice, the author found that while AI can complete tasks, the results may not meet expectations. For example, even if all technical requirements are met, the overall design might not match the author’s vision because his ideas were not clearly articulated and could not be conveyed to AI.
He categorized problems into two types:
- Closed-ended problems: These have clear answers and boundaries (e.g., fixing bugs or following established processes, which AI can handle).
- Open-ended problems: These require creativity and subconscious ideas (e.g., finding ways to significantly improve efficiency or designing a future architecture). The key to solving these problems is to ask the right questions and generate ideas that cannot be expressed in words, as they stem from daily experiences and intuition.
This realization highlighted the value of humans in addressing open-ended issues, where AI can only provide assistance.
4. Two Types of AI-Native Organizations
Based on the proportion of open versus closed-ended problems, AI-Native organizations should adopt two complementary models:
1. AI Control Centers: Designed for creative employees, these centers focus on maintaining their “flow of thought” by providing quick options, validating hypotheses, and handling routine tasks (e.g., researching or drafting initial drafts). For instance, an engineer can use the AI to find similar cases and verify technical feasibility while focusing on creativity.
2. Agent Platforms: These platforms act as scheduling hubs, with AI handling routine tasks and only involving humans when additional human intervention is needed.
These two models can be combined: humans handle open-ended problems and delegate closed-ended tasks to AI platforms for automated execution, allowing them to maintain their focus.
5. Unresolved Questions and an Optimistic Outlook
The author’s current research focuses on how to design AI Control Centers that support creativity and how to make Agent Platforms more effective for handling routine tasks, as well as how to ensure seamless collaboration between the two.
However, he remains optimistic about the future: Humans will not disappear, nor will they become mere tools. AI cannot generate ideas that cannot be expressed in words, which are essential for creative and decision-making processes. The value of humans lies in continuously proposing new problems and generating innovative solutions.
In essence, this article argues that AI is not meant to replace humans but to complement them, with AI handling execution and humans focusing on creativity, leading to more efficient organizational structures.