As AI has moved from chatbots and copilots to agents that can carry out tasks on their own, startups and investors have kept chasing the next narrative. Fovea co-founder Max Jiang argues that the market may be overestimating AI’s short-term impact while underestimating the social and commercial changes needed for the technology to become part of daily life.
In a post on X, Jiang said consumer AI is still at the starting line after several years of startup enthusiasm. The next area worth watching, in his view, may be a shift from the internet’s “attention economy” to an “intent economy” driven by AI agents.
Adoption remains limited despite heavy venture interest
Jiang looked back at the past few years of AI startup cycles. After large language models appeared, the market assumed nearly all content could be regenerated with AI, leading to a wave of chatbot products. When copilots gained traction, founders tried to rebuild productivity software with AI. As AI agents began to show autonomous execution, expectations shifted again, this time toward replacing any human work that could be automated.
His point is that people working at the center of the AI industry can easily overstate how widely the technology has actually spread. He cited a16z’s seventh edition of the Top 100 Gen AI Consumer Apps report, released on Oct. 5. According to that report, nearly half of U.S. consumers have used AI, but only 25% use it every day. As of August this year, only 4.5% of U.S. consumers had an active personal paid subscription to ChatGPT, Gemini, or Claude.
Jiang said that gap shows venture markets and the real world are operating on different timelines. Capital tends to price in long-term potential quickly and reflect it in startup valuations, while broad consumer adoption takes much longer.
He used John Maynard Keynes’ idea of “animal spirits” to explain the pattern. Startups need to show growth before they run out of money. Venture funds need to show portfolio positioning before their next fundraising cycle. Media outlets also tend to emphasize the upside of new technologies. Each participant may be acting rationally in its own interest, but the combined result can still produce collective FOMO.
In his description, the industry has spent too much time focusing on technical infrastructure and too little time asking how long society needs to reorganize itself around new tools. Growth that may take 10 years to materialize gets written into a one-year pitch deck.
From the attention economy to the intent economy
Jiang does not take that as a sign that an AI bubble is about to burst. He sees it as a sign that the market may be approaching an important turning point.
He pointed to the commercial history of the internet. The “attention economy” familiar today was not the internet’s original business model. Early internet services mainly charged for connection time, traffic, and email-related services. Even as the technology improved, the number of people actually using the internet remained limited.
That changed when search engines, social networks, and e-commerce took off. A broader economic structure emerged, linking users, advertisers, merchants, and platforms. Users went online to get information, build relationships, and spend time. Advertisers paid to reach them. Merchants paid for transactions and conversions. Participants that had not previously been directly connected began creating value for one another, and the internet shifted from a technical service into essential infrastructure for everyday life.
Jiang said most AI products today still resemble that earlier stage of the internet. He estimates that more than 80% of AI products still rely on business models built around token sales, seat-based pricing, or subscriptions.
Those models still have value. But if AI is going to reach the consumer market in a deeper way, he said, it may need a new economic relationship that lets different participants create and exchange value together. He calls that possible framework the “intent economy.”
The core idea is this: when the internet reduced information asymmetry, scarce human attention became the carrier of commercial value. If AI reduces capability asymmetry by letting more people complete tasks they could not previously do, then user intent may become the next key entry point for value distribution.
He used shopping as an example. Instead of searching for products, comparing prices, and checking out on their own, users could simply tell a personal AI agent what they want to buy. The agent could find merchants, compare services, and complete the transaction based on the user’s preferences. Merchants, in turn, might pay through transaction commissions or referral-style revenue sharing when the user’s need is successfully met. In that setup, AI services would not necessarily need to charge every consumer a subscription fee. They could earn revenue from helping complete transactions.
The report notes that this transaction scenario is only a potential application of the intent economy. It does not mean Jiang has already presented a complete operating mechanism or a validated business model.
a16z’s October report points in a similar direction. The firm said AI revenue has so far come mainly from subscriptions and usage-based charges, but personal AI agents may have a chance to adopt transaction-led monetization, allowing consumers to access ongoing AI services without paying a fixed subscription.
Owen Chen says many AI apps may really be serving VC demand
Owen Chen offered a more contentious view from the venture side. He said the real demand being served by many AI applications may not be end users at all, but venture capital funds.
He said a friend working at a top AI-native venture firm once raised a basic question: who exactly are the supply side and demand side for the thousands of AI applications now in the market? The intuitive answer is that founders are the supply side and users are the demand side. But in that friend’s view, the demand side that keeps startup supply coming is actually VC.
Chen said he has tried a large number of AI applications himself, but only a handful have become stable parts of his daily workflow. In his view, many AI startups can raise tens of millions of dollars and secure valuations above $100 million, yet still lose their product differentiation and product-market fit quickly when large technology companies update their products or when OpenAI and Anthropic release new features.
He said the market keeps moving from one narrative to another, from general-purpose agents to personal agents, while repeating the same cycle of startup formation, rapid fundraising, and loss of competitive edge.
One reason, he said, may be that venture funds are even more afraid than founders of missing the next major technology trend. For fund managers, failed investments hurt performance. Missing a wave that could reshape an industry can also weaken their ability to raise money from limited partners, or LPs, in the future. That means capital can still flow into products that fit the current narrative even when real market demand has not yet been verified.
This also matches another observation from Jiang. When every company is trying to profit by selling AI capability, model companies start moving downstream into applications, while application companies try to train their own models. The old division of labor becomes less clear, and competition gets tougher. In the end, many companies may struggle to build durable moats, while profits and capital become more concentrated among upstream firms that control core model capabilities.