Stripe co-founder and CEO Patrick Collison said in a recent X thread that personal AI agents could change how information advantages are distributed between businesses and consumers. In his view, the deepest impact may not be how much money these tools save people, but whether they make it easier for genuinely strong products to earn market rewards. Tesla CEO Elon Musk later reposted the thread.
Businesses have long held the upper hand in information and decision-making
Collison opened with a broad idea: companies can, in one sense, be seen as humanity’s earliest form of superintelligence. He was not saying firms possess artificial intelligence beyond human capability. His point was that companies combine large groups of employees, specialized knowledge, data and capital in ways that steadily optimize their own commercial interests.
Consumers, by contrast, often face limited information, limited attention and limited ability to compare every option in front of them. Collison used travel booking as an example. Near the end of a purchase flow, customers are often offered add-on insurance, even though that insurance may cost noticeably more than a policy bought separately. By that stage, the buyer has already spent time completing the booking process and may have little desire to stop and compare alternatives.
He linked that behavior to Herbert Simon’s idea of bounded rationality: people operate under constraints of time, information and cognitive capacity, so they usually do not reach the theoretical best answer in every decision. Collison argued that businesses have long been able to benefit from this asymmetry. If every consumer eventually has an AI agent that can keep searching, comparing and analyzing on their behalf, that balance could start to shift.
Coupons, price discrimination and subscription economics could come under pressure
Collison said some common pricing strategies may weaken if AI agents automatically search for better deals.
He first pointed to price discrimination, where companies charge different prices based on a customer’s willingness to pay, the information they have or the effort they are willing to spend. A product may offer a coupon to shoppers who take the time to look for one, while charging full price to those who do not.
If every consumer’s AI agent automatically finds and applies the best discount code, that form of segmentation becomes harder to sustain. Collison framed it plainly: if every agent seriously searches for coupons, how does the coupon system keep working?
He also said subscription businesses could be affected. Some subscription models, he wrote, effectively rely on cross-subsidies in which customers who forget to cancel keep paying and indirectly support pricing and service for others. If AI agents track usage and remind users to cancel, or help them cancel when the service is no longer needed, that revenue dynamic may become less reliable.
Collison did not say these changes would automatically benefit all consumers. Companies may respond by adjusting prices in other ways, and he said it remains unclear how costs and gains would ultimately be distributed.
The bigger shift, he said, may be that better products stop losing to distribution
Of all the possible effects, Collison said he is most optimistic about what AI agents could do for product-quality competition. Businesses have long faced a basic question: what matters more, the product itself or distribution and marketing power? A startup can build something better than a large incumbent and still fail to get noticed. A company with a strong brand, a large ad budget and established channels can keep selling at scale even without a clear product lead.
He posed a thought experiment: what would markets look like if every consumer effectively had an assistant who spent 10 hours researching a product before each purchase?
In that setting, AI agents could compare specifications, user reviews, after-sales service, price and quality differences instead of steering buyers toward the brand with the most ad exposure. Collison described this as a structural subsidy for product quality. He was not referring to government support. He meant that when the cost of gathering information and comparing products falls, companies that invest in making better products may become easier to discover and better positioned to earn stronger economic returns.
That could also change how businesses allocate resources. Money that once went primarily to advertising and channel control may face stronger pressure to move toward product development and quality improvement. Collison said that outcome would be one of the most valuable economic changes AI agents could bring.
Product reviews and machine-readable information may become more valuable
If AI agents can analyze products in greater depth, then high-quality product information may matter much more than it does today.
Collison said ordinary consumers can only gather and synthesize a limited amount of product data. The internet already contains large volumes of user reviews, professional testing and side-by-side comparisons, but most people are not going to read all of it before every purchase. AI agents can. They can continuously pull from different sources, compare performance, price, reliability and other metrics, then make recommendations based on a user’s needs.
That would make product-quality information itself a more important competitive asset. Collison said many people already believe data is becoming more valuable because it can train AI models. He added a second reason: as intelligence becomes cheaper, high-quality information that AI systems can analyze also becomes more valuable.
In practical terms, companies may need to do more than build good products. They may also need to make sure AI agents can access credible, clear and comparable information about those products. Real-world user experience, third-party testing, after-sales service records and complete specifications could carry more weight in purchase decisions than brand messaging alone.
Search, advertising and traffic-routing models may also change
Collison extended the argument to one of the internet’s core business models: demand routing. Many online companies have built their economics around controlling consumer attention and purchase entry points. A shopper types a product name into a search engine and clicks the first result. A buyer on an e-commerce platform chooses the item ranked near the top. Businesses then spend heavily on ads, search placement and platform fees to secure visibility and traffic.
He argued that this model depends, in part, on the limits of human attention. If AI agents stop following the order of search results and instead search across multiple sources, compare products and complete transactions directly, the commercial value created by placement and exposure could change in a major way.
That could affect search engine optimization, digital advertising and competition among e-commerce platforms. At the same time, Collison did not argue that all paid placement is worthless. A company’s willingness to spend more on advertising can itself act as a market signal, showing consumers which firms are willing to invest to win customers.
Still, if personal AI agents become a new gateway for consumer decisions, the question of who controls traffic allocation, and how much companies must pay to reach buyers, remains open. Collison said he does not yet know the answer, but expects the effect could be large.
Will AI agents make markets more concentrated or more fragmented?
Collison also raised a separate question: will AI agents intensify superstar effects, or make consumer markets more distributed?
One possibility is greater concentration. If AI agents all study products thoroughly and reach similar conclusions, they may decide that a small number of products are clearly better than the rest. In that case, market share could become more concentrated, with a few quality leaders taking a larger portion of demand.
The opposite outcome is also possible. Consumers differ in budget, preferences, needs and use cases. In the past, limited information may have pushed people toward the best-known or easiest-to-buy mainstream products. If AI agents can understand individual preferences in more detail, they may surface niche brands that fit specific users better, pushing markets toward greater fragmentation instead.
Collison used laptop purchases as an example. Different buyers may care more about battery life, weight, repairability, performance or price. AI agents could recommend different products based on those differences rather than steering everyone toward the same popular model.
The key assumption: AI agents must actually work for consumers
Collison said his optimism depends on one central assumption: AI agents must genuinely represent consumer interests.
If their business models rely heavily on advertising, affiliate commissions or paid ranking, then a new question appears. When a user asks for the best product, will the system choose the option that best fits the user’s needs, or the merchant paying the highest commission?
Collison said an AI agent that truly acts on the consumer’s behalf may prove to be the more competitive business model over the long run. But whether that happens is still unresolved.
He closed by describing the market as a large-scale, parallel, decentralized but imperfect machine for assigning value. AI agents, he wrote, could change how that machine distributes benefits to companies. Borrowing the machine-learning term backpropagation, he suggested that if consumers become better at identifying product quality, markets may return sales and profits more quickly to the firms that actually build superior products.
He also acknowledged that this view may be overly optimistic. Even so, he said he hopes companies that take pride in their products and seriously invest in research, development and quality improvement will eventually feel that the market is moving in their favor.