User-side software, and adversarial marketplaces

Two questions I've been asking myself as I brainstorm my next moves are what truths will still hold 5 or 10 years down the line, and where will language models change cost structures enough to create new opportunities. Unexpectedly, I keep landing on consumer ideas, especially in fragmented two-sided marketplaces compromised by the platforms' incentives and principal-agent problems. In both, LLMs could reduce search and coordination costs. Regardless of their marketing, platforms don’t always help us find the best option. Often, they nudge us toward what's most profitable for them. Sometimes they do it by exploiting information asymmetries. Other times, by refusing to expose a filter they obviously have in structured data, dressing up ads as native recommendations, or simply nudging us to a higher fee product. LLMs might reduce the cost of high-variance coordination work that used to require humans. AI changes the dynamic when it's not implemented as a concierge layer owned by a…