People Layer Models
Cursor's $60 Billion Exit Rewrites AI's Value Map — and ChatGPT Logs Are Showing Up in Court
The acquisition of Cursor, the AI coding assistant, at a valuation of sixty billion dollars has given venture firm Andreessen Horowitz a concrete exhibit for a thesis it has been building: the value in the AI ecosystem is not concentrating in foundational models, but in the interface layer — the specialized tools users actually touch to accomplish specific tasks. The business logic is grounded in observable market dynamics. As multiple capable foundational models have become available, API access prices have dropped significantly, and the raw capability gap between leading models has narrowed, what differentiates user experience is increasingly the layer above the model: workflow integration, contextual specificity, interaction design. Cursor built specifically for software developers, embedding deeply into coding workflows in ways that generated the kind of user stickiness that translates to sixty billion dollars in exit value — more than many software companies that have been operating for decades.
The legal system is simultaneously surfacing a consequence of AI's proliferation that most users have not considered: private conversations with AI chatbots carry no legal privilege whatsoever. No attorney-client protection, no doctor-patient confidentiality — just data, fully subpoenable in both civil and criminal proceedings. Lawyers are increasingly requesting those logs, and the cases in which they are appearing as evidence are multiplying. The specific danger is behavioral: the conversational interface of AI chatbots creates a psychological sense of intimacy and privacy that has led people to disclose things they might not commit to any other medium. That subjective sense of privacy has no legal analog.
The chilling effect, if this becomes widely understood, could be significant. Users who know their AI conversations are legally discoverable may become more guarded in how they use these tools — potentially reducing some of the genuine utility people have found in them for working through difficult personal decisions and problems. The pattern is not entirely new — text messages, emails, and social media direct messages have been legally discoverable for years — but the nature of what people tend to confide to AI systems makes the exposure qualitatively different.
The Cursor exit also poses a question for antitrust regulators that will become increasingly pressing: if high-value interface companies end up absorbed into larger ecosystems that have monopoly characteristics at the infrastructure level, do the structural leverage concerns of antitrust law come into play? Under the Sherman Antitrust Act of 1890 — still the governing statute — having a large market share is not itself illegal. What the law prohibits is maintaining a monopoly through exclusionary conduct: actively blocking competitors from reaching users rather than simply outcompeting them. For AI, the relevant regulatory question is whether control over one layer, such as cloud infrastructure, is being used to disadvantage competitors at another, such as model deployment or interface distribution. Market share is the starting point of the inquiry, not its conclusion.