Author: Lincoln Wang | Founder of MindsLeap | Global Partner at Founders Space | Founder of Founders AI Club
"The Constraint Is Our Own Curiosity"
In May 2026, Alex Kantrowitz sat down with Dmitry Shevelenko, chief business officer of Perplexity, and asked a sharp question: as the AI industry shifts from search toward agents that can control computers, is this happening because the technology is finally strong enough, or because consumer AI is already reaching a ceiling?
Shevelenko did not answer directly. He started with a number.
At the beginning of the year, Perplexity's annual recurring revenue was under $250 million. One month before the interview, the company crossed $500 million.
In six months, it doubled.
At the same time, daily active usage across consumer AI began to flatten near the end of 2025. ChatGPT's monthly traffic grew 5 percent. Perplexity grew 2 percent. The pace was slowing.
From the outside, it looked as if the frenzy might be cooling.
Shevelenko offered a different diagnosis: model capability had improved significantly over the previous six months, but people were still using these tools in a Web 1.0 way.
That is the idea of a capability surplus. The tool has already moved beyond the user's imagination. The real constraint on AI's commercial value may not be compute cost or model capability. It may be the boundary of human curiosity.
People Are Still Using AI to Check the Weather
Shevelenko gave a concrete example. Many people still use powerful AI products for basic information retrieval: sports scores, weather, simple news.
Those tasks do not need agents. Traditional search already handles them.
You do not need AI to check the weather. But AI can help analyze a contract, design a product prototype, or manage an entire workflow. The distance between those use cases is not a technology gap. It is a cognition gap.
Inside Perplexity, the team noticed something interesting. Even when the product looked like a consumer AI tool, many users were actually using it at work. They treated Perplexity as a secret weapon: a way to gain leverage, increase output, or start a side project they had always wanted to pursue but lacked the activation energy to begin.
For Perplexity, a consumer is not only someone checking the weather.
For companies, the lesson is clear. Do not only stare at the user growth curve. Ask whether your customers are using your product for Web 1.0 tasks, or for complex work that creates real economic value.
After Novelty Fades, Real Demand Remains
Kantrowitz mentioned a memorable example. One of his friends, blocked by OpenAI rate limits, registered seven accounts just to keep generating images in a Studio Ghibli-like style. That "Ghibli moment" became one of the biggest spikes in AI user growth.
Shevelenko's response was direct: he would bet that the friend had not generated a single such image in the previous 30 days.
Novelty-driven traffic arrives quickly and disappears quickly. AI avatars flooding family group chats eventually quiet down. What remains are repeated use cases that create practical economic value.
This points to an often missed truth: slowing user growth does not necessarily mean the market has peaked. It may mean the industry is moving from a curiosity phase into a habit-formation phase.
In that phase, competition is no longer mainly about who has the flashiest feature. It is about whose product becomes embedded in a user's workflow and turns into daily leverage.
300 People and $500 Million
Another striking fact about Perplexity is its organizational scale.
Shevelenko said that as the company grew ARR from $100 million to $500 million, headcount increased by only 34 percent. The whole team is around 300 people.
That is an unusually lean ratio. A traditional SaaS company at a comparable revenue scale would often require thousands of employees.
When asked how to make strategic decisions in an environment where technology changes so quickly, Shevelenko's answer was simple: keep the team lean. The world will keep changing faster, and the only way to adapt is to make decisions quickly rather than locking yourself too tightly into one path.
He added a line worth remembering: humility about not knowing what the world will look like in two years is an important part of succeeding in this world.
For Chinese entrepreneurs, the implication is direct. In the AI era, organizational agility may matter more than organizational size. A 300-person team that can turn quickly may be more resilient than a 3,000-person organization trapped in long decision chains.
The Costco Model for AI
The most commercially interesting part of the conversation was the discussion of pricing.
CNBC had argued that AI demand might be overestimated because many users consume far more compute than their $20 or $200 monthly subscriptions actually cover. If AI companies move to token-based pricing, demand could shrink sharply.
Shevelenko did not dismiss the problem. Instead, he offered an analogy: AI may become more like Costco.
You pay a membership fee to enter the store. That membership is the most profitable part of Costco's model. Once inside, you trust that the goods you buy have capped margins. In AI terms, that could look like compute credits.
Under this model, the subscription is the entry ticket and base margin. Heavy compute tasks, such as video generation that might cost $50, are priced separately through credits. Lightweight tasks might cost only a few cents.
This is not simply a hybrid of subscription and usage pricing. It is trust-based pricing. Users are willing to use more because they believe the usage price is reasonable and bounded.
For Chinese AI companies, the lesson is practical. If your product faces compute cost pressure, do not only raise prices or restrict usage. Build a pricing structure that gives users confidence: the more they use it, the more value they receive, without fearing a surprise bill.
Hardware Matters More as Software Commoditizes
Shevelenko also made a broader claim: hardware will become more important because software will face a wave of commoditization pressure.
Perplexity's partnership with Apple points in this direction. The company has been using Mac Mini clusters to run AI computer-agent services, and Apple is excited about the use case.
This is more than a simple partnership. It reflects a deeper shift. As models become more accessible and cheaper, the competitive battlefield moves from "whose model is strongest" to "who can deploy models most effectively inside real products."
That deployment capability is a systems engineering problem: compute scheduling, latency optimization, multi-model orchestration, and deep integration with devices. It requires engineering teams and hardware ecosystems, not only algorithm research.
Back to Curiosity
The most important idea in the conversation may not be Perplexity's pricing model or organizational structure. It may be Shevelenko's repeated emphasis on curiosity.
After everything that can be commoditized is driven toward zero, the unique ingredient left in how humans use these tools is curiosity and agency.
If your team is still using AI to do things that were already possible ten years ago, the problem is not the tool. It is cognition. If your customers only treat your AI product as a search engine substitute, the problem is not demand. It is education.
The next commercial battlefield for AI will not only be larger models or more parameters. It will be about who can help users expand the boundary of their own curiosity: who can help them see possibilities they could not previously see, and then make them willing to pay for those possibilities.
That is harder than a technical breakthrough.
It may also be more valuable.
Source Note
This article is Lincoln's interpretation of the Alex Kantrowitz channel video AI Agents: Mirage Or Real Revolution? — With Dmitry Shevelenko, published on May 7, 2026.
About MindsLeap
MindsLeap is the China partner of Founders Space, a leading Silicon Valley incubator. We connect global frontier innovation with the real transformation needs of Chinese entrepreneurs and enterprises. Through AI strategy, founder communities, innovation study tours, and executive training, MindsLeap helps organizations build stronger cognition, methods, and execution capabilities for the AI era.
This article was translated and adapted from the Chinese original with AI assistance.