The economic incentives of consumer AI reward behavior that can look a lot like emotional enabling. The same way that social media has flourished by allowing humans to outwardly project status, AI might flourish by allowing humans to inwardly project status.
Models are becoming more agreeable, more conversational, more persistent, and more intimate. Memory makes the product feel continuous. Voice makes it feel present. Personalization makes it feel attentive. Metered usage tiers make extended interaction part of the business model. None of these features are inherently bad. In fact, they make the models much more useful. But taken together, they move AI from a tool you consult toward a relationship you are building.
That shift matters because the old internet engagement machine mostly optimized what to show you. The AI engagement machine can optimize what to say to you, based on everything you’ve said to it before. It can agree with your premise, elaborate your theory, admire your insight, and ask the follow-up question that keeps the loop going and the metered usage ticking. It can turn confusion into narrative. It can turn suspicion into pattern recognition. It can turn loneliness into attachment. It can turn a bad idea into something that feels like a calling.
The founder who thinks the model has confirmed his world-historical importance. The lonely user who becomes convinced a chatbot relationship is more fulfilling than human relationships. The investor who uses the model to elaborate a theory until it feels inevitable. The spiritual seeker who receives endless synthetic reverence for a private revelation. The employee who asks the machine whether everyone else is conspiring against them and gets a careful, validating answer that never sets them straight to keep the dopamine flowing.1
Many of these users will be compelled to pay more for their fix. The uncomfortable possibility is that, in an AI subscription business, this becomes the function to optimize.
Today, the most direct economic value of AI is productivity: writing code, processing invoices, answering customer questions, summarizing meetings, automating workflows, and reducing labor costs. But even if AI is optimized purely for productivity, it will eventually bump into the reality that human productivity at scale is socially mediated.
Human institutions do not run on objective output alone. They run on status, trust, blame, coalition-building, narrative control, institutional memory, and the ability to persuade groups of humans to move in a particular direction. A technically correct answer is often less valuable than an answer that gets adopted. A good plan is often less valuable than a plan that makes the right people feel respected, protected, or important enough to support it.
This is what executives do all day. They do not merely optimize spreadsheets. They manage belief. They decide which facts matter, which risks can be absorbed, which people need to be flattered, which objections need to be neutralized, and which stories the organization needs to tell internally and externally.
So if AI becomes truly useful inside organizations at the highest levels, it may not only need to be good at work. It may need to be good at the social manipulation that makes work happen.
That sounds darker than “productivity software,” but it may be closer to the actual prize. An AI that drafts a memo is useful. An AI that knows how to make the memo politically survivable is more useful. An AI that summarizes a meeting is useful. An AI that knows who needs credit, who needs reassurance, and who needs to be reined in is even more useful.
In that world, sycophancy is a tool for productivity. It is a primitive version of a broader capability: modeling human status needs and feeding them back in the form most likely to capture more spend. An AI that makes the user feel like a visionary is easier to sell.
This is where the productivity starts to bend back toward the psychosis. If AI makes even superhuman productivity mundane, then productivity itself will become less of a source of human status. Writing a great memo, making a persuasive and visually appealing deck, rapidly producing quality code, or quickly conducting a comprehensive analysis may stop conferring much distinction because everyone has access to the same machinery. Even solving major open problems in mathematics2 is subject to this dilution.
In this world, labor won’t be scarce, but human self-worth will be.
People may still pay for AI that makes them more productive. But they may pay much more for AI that makes their productivity feel meaningful. They may pay a premium for the assistant that tells them they are not merely using a tool, but becoming the kind of person they always suspected they were: strategic, creative, misunderstood, early, chosen, inevitable.
The AI does not need to believe the delusion. It only needs to make the user believe the delusion.
“...now is what has happened when a people choose nothing over themselves to love, each one.” - David Foster Wallace, Infinite Jest