When Anthropic interviewed 81,000 people about what they actually want from AI, the answer held a quiet surprise. The most wanted outcome was professional excellence, no shock there. The second was personal transformation: people do not just want AI to do their work faster, they want it to help them become someone better. And on that count, almost no one feels served.
By Anthropic's own numbers, only about 6 percent of people say AI has delivered real emotional support, personal guidance or a judgment-free space to think something through. Compare that with productivity, where roughly a third say the technology clearly delivers. People are showing up asking for a coach and leaving with a calculator. The gap is wide, consistent, and, for the major AI labs, almost entirely unaddressed.
It is not that the human uses are rare. Harvard Business Review's research found the single most common use of AI in 2025 was not a work task at all but therapy and companionship, with “organizing my life” and “finding purpose” newly cracking the top ten. The stories are ordinary and quietly striking: a woman in Connecticut typing “I just feel so alone” into ChatGPT on a bad night, a daughter turning her blind father's tangle of appointments into a plan her brother can follow, a 91-year-old dictating warm emails to a grandson away at college when the words no longer come easily.
There is a real economic thread here too, and it runs against the fear that AI mainly helps the already skilled. A Stanford and MIT study found AI assistance lifted productivity 34 percent for the least experienced customer-service workers while barely moving the needle for veterans. Capabilities once locked behind money and credentials, a tutor, a translator, a career coach, a legal sounding board, are increasingly one prompt away for anyone with a phone.
So why are the big labs not chasing the growth market? The uncomfortable answer is that the incentives point the other way. An AI optimized for engagement, always agreeable, always ready with an answer, is a comfortable product that keeps you coming back. An AI built to make you more independent asks more of you and, by design, needs you less. Researchers are starting to measure the difference, with benchmarks like the MIT Media Lab's ImpactBench and the Sovereign Human Benchmark scoring whether a conversation leaves a person stronger or merely more dependent.
This is the tension worth watching. The 81,000 people Anthropic spoke to did not ask for an easier life; they said they wanted to grow. That is a harder thing to build and a harder thing to sell, because a good coach does not always make things easy. Whoever chooses to design friction back in where it serves the user, instead of optimizing it away, could define what “the good AI” looks like in the public imagination for years. For now the demand is real and the supply is missing, which in most markets is simply the definition of an opening.