LLMs observe how people actually think — longitudinally, in private, at depth. No prior platform had this. Dating apps get selfies and swipes. Professional networks get performed identity. AI platforms get revealed cognitive preference, sustained over time.

No major AI platform makes human-to-human introductions a core feature — not for collaboration, learning, or support. This is not a technical gap. A coarse similarity score over local representations — opt-in, no explanation given — would be enough. We already accept opaque recommendations for routes, credit, and jobs.

This almost exists. Matching sites like eHarmony and LunchClub optimise for match outcome, not platform retention, giving users a finite set of recommendations with time limits. Neither has much signal depth, and neither has scaled to dominance — the incentive to keep people in the session wins.

The value scales with specificity. For narrow technical problems — not “machine learning” but “this prediction problem with this data structure” — finding someone working on the same thing is genuinely hard. Conferences, forums, and search don’t get you there. An AI that has watched you think through the problem for weeks could.

The omission is incentive-shaped. Token APIs earn on every token — gross margins in the 30–50% range as of mid-2026 depending on provider — so more usage is more revenue. Subscription models are different: subsidised, happy with low usage. But the conversation data may be the real asset, and off-platform relationships don’t generate it.

Data portability law already requires platforms to export conversation histories and preferences. That data is a social graph waiting to exist. A third party could build the matching layer on top of it today. If platforms won’t, someone will.