Personal AI agents could reward better products, writes Patrick Collison

In an X post on 9 October 2026, Patrick CollisonDealroom has a profile for this one. Try Dealroom →, co-founder and CEO of Stripe, set out how he thinks personal AI agents – software that shops, compares and decides on a person's behalf – could change the way markets work. In plain terms, his main points were:

• Shoppers get a smarter helper: Companies have long had more time, data and brainpower than the customers they sell to, and some profit from people being distracted or underinformed (his example: the overpriced travel insurance offered at the end of a booking). Agents working for the consumer should make people more rational buyers.
• Some old tricks stop working: Coupons and other ways of charging different people different prices work less well if every agent hunts for the best deal. People who forget to cancel subscriptions would stop quietly subsidising everyone else. He says it is unclear who wins and loses overall.
• Quality beats marketing: Today a better product can lose to a better-known one. If every purchase is researched as if someone spent ten hours on it, making the product better becomes a more effective strategy than winning on distribution. He calls this good news.
• Trustworthy product information becomes more valuable: Agents will hungrily collect signals about quality, so data matters not only for training AI models but as fuel for cheaper intelligence.
• One winner or more variety? Agents could push everyone towards the same top product, or instead match each person to their own tastes and break up today's artificial clustering. He leaves this open.
• Search and ads may be upended: Many internet businesses earn money by routing short attention spans to whoever ranks first or pays for placement. How this changes when agents do the looking is unknown, but he expects the effects could be large.
• A key assumption: All of this depends on agents genuinely acting for the consumer, which he suspects will be the winning approach but which raises many further questions.

His conclusion: the market is a huge, imperfect machine for rewarding value, and agents may tilt it so that companies making things they are proud of are rewarded more. He admits this may be optimistic. The view fits Stripe's own push into agentic commerce, the payments rails that let AI agents buy on users' behalf.

Read more: Patrick Collison on X