Serhiy Tokarev Explains Why AI Agents Will Not Save Business Without Process Restructuring

Autonomous AI agents are already conducting audits, negotiating deals, and managing operations. Because of this, it may seem that the corporate world has entered a new era. And this is not just short-term hype: artificial intelligence (AI) is genuinely changing the rules of doing business.

A gap is already visible between companies that are gradually integrating AI into their processes and those that still see it as an off-the-shelf solution. Serhiy Tokarev, an investor and founder of the Tokarev Foundation, explained why AI is becoming essential for business and how it should be implemented to deliver real results.

AI Should Understand Information but Not Always Act Independently

When analysing a startup, AI can become the first analyst. At Roosh, an internal assistant built on Claude ranks a long list by team excellence and market opportunity, returning a short rationale for each company. Analysts begin deeper research with the top 10 instead of checking 100 startups one by one; productivity in initial screening has increased roughly tenfold.

AI takes over the collection of basic facts and saves a significant amount of time. At the same time, Serhiy Tokarev does not rely on the model to decide where to invest. In his view, the final decision, signature, and responsibility must remain with a human.

“I doubt that human oversight of AI is only a temporary stage that will become unnecessary once this technology becomes more accurate and intelligent. According to Deloitte, 47% of corporate users have made decisions based on incorrect model outputs, while 77% of business leaders are concerned about risks associated with generative artificial intelligence. That is why neither full control nor responsibility can be handed over to an algorithm,” notes the founder of the Tokarev Foundation.

The transition of a model from reading information and preparing recommendations to taking independent action must happen only within clearly defined boundaries. Human oversight does not hinder AI. On the contrary, it is what makes the technology suitable for business.

Artificial intelligence is only one component of the entire process. Roosh’s NDA agent, for example, follows a handbook of acceptable terms, proposed changes and issues to escalate. It prepares an initial review and redline that a lawyer checks and refines, reducing processing time from about 1.5 hours to 15–20 minutes. The first deployment also showed why review matters: the agent sometimes rewrote a provision whose different wording was already acceptable. Adding more examples of valid language helped correct this.

Processes Win, Not Models

Serhiy Tokarev is convinced that access to a frontier AI model should not be considered a long-term competitive advantage. Real value comes from deep integration, evaluation systems, and proprietary industry data. Roosh’s unified LP platform built with Claude Code illustrates this: it brings fund materials and updates from more than 60 portfolio companies into one current context, reducing the time spent assembling and reconciling versions for each prospective LP conversation.

“In our portfolio companies, models do not replace teams of specialists — they help them and expand their capabilities. Some AI agents review and analyse information flows, while others synthesise research in fintech, medical technologies, and game development. Their value comes from how they work together,” says the investor.

The question is not whether AI will be able to replace people. The real question is where exactly the model is already creating measurable business value, and where the market is still selling the future instead of the present.