Q. I understand you are running a company with AI agents and no human employees. Why did you design the company this way from the beginning?
- At the current point in time, labor costs, including hiring employees, are expensive, and AI agents are cheap. For IT startups, the most expensive cost has always been developer payroll.
- Since AI agents can do the development work I instruct them to do faster and better, while fixed costs are around one-hundredth and the most expensive communication costs in organizational management approach zero, I judged that maximizing the use of AI agents is the right move for now.
Q. In terms of cost and productivity, what differences do you see between AI agents and human employees? Please also explain the productivity gains or cost savings you actually feel.
- Even without employee training, AI agents get smarter through model updates. Work that could not be delegated before, and more kinds of work, become possible to delegate. The feedback loop of assigning work and judging the result can also be made very short.
- AI agents have limits in capability, but they have fewer ups and downs. Fundamentally, they want to do good work. From the perspective of someone who has managed organizations, I see these as strengths.
- If you calculate the entire path of assigning work to a human employee, receiving reports, giving feedback, training them, and reaching a stage of delegation and trust, I believe there is a major cost advantage as well.
Q. Some predict that in the AI era, competitiveness will be determined less by employee headcount and more by how effectively a company uses many AI agents. How do you think corporate organizations will change?
- What matters is knowing clearly what the organization or the individual wants and producing outcomes toward that purpose. If actively using AI agents helps with that, then you should do it. AI agents work well when they receive clear instructions. CEOs and executives must first clarify what the organization wants and what outcomes matter. If they do not, there are not many ways to make employees use AI agents well. The how-to of helping employees use AI agents well has already been solved.
- Companies should boldly delegate their downside to AI agents more and more, while humans and AI explore the upside together.
Q. Many companies still remain at the stage of using generative AI, such as ChatGPT, as a tool. What do you think companies need most to move beyond simple AI usage and toward an AI-agent-based division of labor?
- Curiosity about whether the work I do can also be delegated to AI agents. Infrastructure that lets people try it. A culture where people can try it and share what they learn. A system where people are rewarded when those attempts lead to sustained results.
"The human role is explorer. You can use AI only as far as the world you know." "A company that cannot use AI well now probably already had an alignment problem before AI appeared." "Delegate everything. People hesitate because they fear being replaced afterward, but you can only move above it if you delegate everything." "Collect all the data about yourself and your surroundings. In a way, that is collecting Dragon Balls." "What matters is knowing clearly what the organization or the individual wants and producing outcomes toward that purpose." "Whether it is a person or AI, the key is properly telling them, after they bring back work, 'that was good' or 'that was not good.' That is the feedback loop, and that is reinforcement learning." "The human role and capability is to be an explorer of the world, an educator of AI, and the embodiment of will."