Perplexity

Perplexity serves on ToTheMoon as a running test of whether a company with no model of its own can build a business on search. The hosts have come back to it since 2024 when discussing the unit economics of AI search, its browser, agents acting for a user, and the cost of an error when such results end up in serious documents.

The first take was skeptical: running a large model on every query is expensive, the economics are papered over with investor money as at Uber, and the risk is greater because big companies build search (episode 007). The Pages feature was summed up as bolting Notion onto Perplexity, and the startup's opening was seen in Google's caution (episode 009). In autumn 2024 the hosts set its eight-billion valuation against OpenAI's hundred and sixty (episode 031). In 2025 its browser was read as a bid for direct access to the user (episode 056), while inside Perplexity you cannot run O1 Pro or Deep Research, because it depends on other people's models (episode 050).

Later episodes moved the subject to agents and accountability. The Amazon dispute, in which an agent enters a store as if it were a person and buys on their behalf, was framed as a question of identification and commissions (episode 083); by 2026 the issue was authorization (episode 147). A judge's clerk who used Perplexity and put invented facts into a filing became, in episode 157, the example of false confidence. By September 2026 the verdict was blunt: reselling other companies' models is a dead business, and Anthropic's open memory will not work through someone else's interface (episode 161).