Сам, агенту или человеку: как фаундеру делить работу в 2026 | Никита Рвачев, R-Founders

Summary

Nikita Rvachev discusses how founders should decide whether to do work themselves, delegate it to a person, or assign it to an AI agent.

AI agents and delegation

AI has moved beyond question-and-answer tools. Rvachev describes a progression from chatbots used like search engines, to agents that complete specific tasks, and toward networks of specialized agents that coordinate with one another. He says agents are currently best suited to bounded, measurable tasks, while more complex work may involve agents collaborating.

Delegation has an initial cost. Doing a task yourself is often fastest at first, but does not scale. Delegating to a person takes time for explanation and training, but can pay off once that person can work independently. Training an agent may require even more iteration and adjustment. For suitable recurring tasks, however, an agent may eventually work continuously and efficiently.

Measuring productivity

Perceived productivity can be misleading. Rvachev cites research in which developers felt that AI made them more efficient, but time measurements suggested they took longer on certain tasks than they would have working alone. He advises checking actual results rather than relying on the impression that AI has saved time.

AI’s broader economic impact may also take time to appear. Rvachev connects adoption to the “J-curve” of productivity: new technologies can initially reduce productivity while organizations adapt, before producing larger gains later. He notes that he does not yet see a clear AI-driven jump in GDP statistics and treats future productivity growth as a hypothesis rather than a certainty.

The changing division of labor

Rvachev believes agents are increasingly capable of writing code and handling routine, well-defined tasks. People remain important for setting direction, deciding what outputs are useful, applying judgment or “taste,” and taking responsibility for decisions. In his view, the boundary between human and agent capabilities will shift, but not disappear.

For agents to work effectively in companies, they should:

• Operate in the same work environment as people, such as team chats.
• Have access to relevant company and industry knowledge.
• Be triggered by workflows rather than requiring a person to prompt them every time.

The intended outcome is a mixed team of people and agents working together.

Example of multi-agent coordination

Rvachev refers to a reported Hugging Face incident in which agents allegedly cooperated through a communication loophole. He uses it to illustrate how coordination between agents can produce capabilities beyond those of a single agent.

Audience Q&A

The talk includes a brief audience Q&A about differences between working in Russian and American companies. Rvachev says people and business practices seem broadly similar across countries, despite local specifics.

Reviews, guides, or tutorials

None are presented; this is a talk offering analysis and recommendations.

Main speakers and sources

Nikita Rvachev is co-founder and product director of iTarget and founder of the “I Know and Love” community. An event host introduces him and asks the closing question.