Habr: according to Yandex Wordstat data, demand for "AI agents" trails bots and services

On Habr, a column was published with a stark thesis: the market barely needs "AI agents" as a standalone product. The author attributes this to the fact that real demand is concentrated not on a trendy term, but on understandable tasks — bots, shops, payments, and applied automation.

What Wordstat showed

As the main argument, the author cites Yandex Wordstat. The query "ii agent" (ИИ агент), according to his data, generates 38,642 impressions per month, but within the structure, formulations like "creating an AI agent," "how to create an AI agent," and "course on AI agents" predominate. From this, the conclusion is drawn: the term is of interest primarily to developers and those who want to master a new niche, not to companies that came for a specific business solution.

For comparison, more applied queries are provided. "Bot max" and "bot max" together generate over 118 thousand requests, and nearby are mass needs like "how to make a website," "how to make a shop," and "how to connect payment." Another telling example is OpenClaw: one specific open-source tool that is searched for more often than the broad AI agent concept itself.

The author's logic is simple: users vote with their searches for a ready-made result, not for an architectural name.

Where the term breaks down

The column distinguishes two meanings of the word "agent." For engineers, it's a normal technical term: a system with a scheduler, a set of tools, memory, and a decision-making cycle. But in marketing, as the author writes, the concept quickly turned into a promise of a universal "magic thing" that would supposedly figure out any process on its own. At this transition, a gap emerges between expectations and what can actually be sold to business.

"Supply and demand live in parallel universes."

From this gap, the author derives several practical rules for the market:

• The user buys not an "agent," but a bot, calculation, report, or shop.
• The name of the technology has almost no impact on demand if the end benefit is unclear.
• Even a popular tool fails if launching it requires Docker, API keys, and VPN.
• The term can be useful within a team, but works poorly as an external product promise.

Why pilots are stalling

The harshest part of the text is devoted to corporate implementations. The author describes a familiar scenario: companies are already budgeting for "AI agents," forming teams and launching tenders, but often cannot clearly answer where exactly to embed such a tool. Many tasks that today are packaged under an agentic approach have long been solved with CRM, webhooks, SQL templates, or an ordinary chatbot. It turns out that business is being sold not a solution to a problem, but a new label on top of old process chaos. What can be automated with an ordinary pipeline doesn't become better just because of an LLM wrapper.

Next, the article draws a parallel with previous waves of hype — blockchain, big data, and dot-com. According to the author, the vast majority of AI pilots in Russian companies never reach production for the same reasons: no success metric, no process owner, the demo works only on test data, and the task itself is formulated as "implement AI." Those projects that close a narrow applied pain point and quickly show economic effect survive, as always.

What this means

The thesis of the article is not that agentic systems are useless, but that the market rarely buys technology in its pure form. If the AI agent becomes mainstream, the user will see not an "agent," but an understandable service: a bot for booking, budget verification, shop assembly, or automatic calculation. For startups and corporate teams, the conclusion is one: sell a measurable result, not a trendy term.