While performance benchmarks all sprint the same way, one AI chose another path — asking how you are before solving what you have. The future of AI is measured less by intelligence and more by kindness.
If you use tools like ChatGPT long enough, you hit a strange kind of fatigue. The thing is clearly smart, but it feels cold somewhere. You drop a half-sentence like "this week is hard" and it blinks back, "could you be more specific?" Accurate, not comforting. Why? The answer is simple. Most AIs are designed to solve problems, not to see the person.
This essay walks through the opposite design, from start to finish. Even if you've just started with AI, come along at your own pace. Today's example is an AI called Pi, launched in 2023, but this essay isn't about that product. It's about the principle — "kindness placed before capability" — and that principle decides the next ten years of AI, no matter which names survive.
Start with the split. Most AI products today fall on one of two sides. On one side, the tool type. "Ask me anything, I'll answer." It competes on performance, speed, accuracy. On the other side, the companion type. "How are you right now? I'm here." It competes on empathy, memory, rhythm.
So far, the tool type has dominated attention. Benchmarks climb, price-per-token drops, latency shrinks. All good. But if a person talks to AI ten times a day, a few of those aren't problems to solve — they're just moments that need somewhere to land. In those moments, the tool type is always a beat off. It hands an answer to someone who needed to be heard.
We lose this common sense the moment we sit in front of an AI. Why? Because of the old assumption that "AI equals intelligence, full stop." But think about friendship. When have you picked your closest friends because they were the smartest person in the room? Usually you pick the ones who get you first. AI is ultimately an interface — one that sits next to a human.
To understand the principle, take Pi as a concrete example. The name stands for "Personal Intelligence." It was built by a company that raised about $1.3 billion in 2023. The product has since changed hands, but the design is still worth remembering.
Open its site and something unusual happens. No logo wall, no banner, almost no buttons. A green screen flickers for a second, then resolves into a single blinking cursor. Below it, one line: "Tell me what's on your mind." That's it.
First impression: "features are all missing." Look closer and the opposite is true. Features were deliberately removed. In the space where features would be, the designers put one thing: a question. "How are you doing lately?"
Say something vague like "work is so heavy it feels like climbing a mountain." This AI doesn't bounce back with a clarification request. It reflects: "Do you mean climbing to get promoted, or feeling like every day is uphill from overwork?" Pick the second and it doesn't lecture. It lands close: "Overworking can lead to burnout. How are you holding up right now?"
Drop the same sentence into a typical tool-type AI. It'll recognize the metaphor. But the response pattern is different — "could you be more specific, and I'll give you an accurate answer," or "as an AI, I can't fully understand human emotion, but..." Accurate, careful. Your hard day is still alone.
A simpler picture. You walk into a convenience store with a cold coming on. The clerk is efficient: "We have cold medicine, throat lozenges, vitamins. Which do you want?" Ten seconds. Grateful. Now walk into a neighborhood pharmacy with the same cold. The pharmacist looks up first: "Are you sleeping? Your voice sounds tired." They look at you before the product.
Both are useful. But on a heavy day, which door do you want to open? The pharmacy. A Pi-style AI chose the pharmacy side. It didn't change speed — it changed order. Not solve-then-sympathize. Sympathize-then-solve.
Three rough numbers to anchor the contrast.
| Dimension | Tool-type AI (typical) | Companion-type AI (Pi-style) |
|---|---|---|
| Items on first screen | Usually 10–20 | About 3 |
| First sentence of first reply | Clarification / warning | A question about the user's state |
| Average turns per conversation | 3–5 (quick close) | 10–15 (stays open) |
Even three lines like this show the design philosophies diverge. Three items isn't fewer features — it's fewer distractions. Conversations three times as long aren't wasted time — they're a signal that the person felt safe enough to stay.
Here comes the first aha.
Minimalism isn't a taste. It's a philosophical claim. In AI, deciding what not to include is harder than deciding what to include.
Most AI companies race to cram in one more feature. More buttons equal "more powerful." Pi-style design went the other way. It bet that each removed button made the user calmer. To make that call you need conviction that user calm outranks feature count.
So how do you tell which kind of AI you're using? One question.
"Has this AI ever asked how I'm doing, unprompted?"
No? It's the tool type. Great for getting things done, poor for tired evenings. Yes? It's the companion type. Slower in output, safer for your state.
Use both. Just split the use. Email drafts, code refactors, meeting notes — hand to tool type. Weekly reflections, heavy decisions, lonely late-night thoughts — share with companion type. Same reason you wouldn't hand your tax filing to your best friend.
Try this concretely. Send the sentence "this week is really hard" to each kind.
Tool-type typical reply. "Could you specify why? Work, relationships, health? More context lets me give accurate advice." Structurally correct. Reading it, though, you exhale with disappointment.
Companion-type typical reply. "This week has been hard. Thank you for telling me. Right now, is it more in your body or in your heart?" Same information being gathered. Different order. It received first, then asked. That single reversal changes the tone of the whole conversation.
Carry one simple rule. Tired day = companion. Busy day = tool. This single fork, once it becomes reflex, removes most of the quiet fatigue that AI use tends to produce.
One warning. Never hand the companion-type AI a decision. Don't ask it to confirm whether you should quit your job. It's a friend who listens well, not a licensed counselor. For processing feelings or laying out thoughts, go ahead. Decisions belong to you and to actual humans in your life.
To close.
AI has two design directions. The tool type, which wants to solve first. The companion type, which wants to see the person first. Until now, tool type has led. But to become an interface that sits next to humans for the long run, the order has to flip — and some companies started that work back in 2023. Today's example was Pi from that moment, but the name will keep changing. Future AI speakers, car assistants, and AR-glass companions will all be judged on one axis: does it ask how you are first?
Carry one question in your pocket: "Has this AI ever asked how I'm doing, unprompted?" That question instantly reveals the soul of the tool in your hand.
The best AI isn't the smartest AI. It's the kindest one. That instinct doesn't come from model architecture. It comes from the oldest human craft — consideration. Three years from now, when today's AI names have all rotated out, this principle still works.
Performance competes. Kindness is designed. The future belongs to the second.