I built an AI life coach — here's the uncomfortable truth about what people actually need

I've been watching thousands of coaching conversations flow through Coach4Life — the AI coaching platform I built. And one pattern keeps appearing that nobody talks about.## Nobody actually wants adviceSeriously. When someone comes to a coach and says "I don't know what to do," they usually already know. They want someone to help them hear themselves say it.I tested this theory by analyzing sessions where users got direct advice versus sessions where the AI asked clarifying questions. The outcome difference was striking:- Direct advice sessions: ~12% action rate (user follows through)- Question-based sessions: ~41% action rateThe insight has to come from inside the person. Any coach — human or AI — who misses this spends most of their time giving advice that never gets acted on.## The moment I rewrote everythingThe first version of Coach4Life was... basically a fancy chatbot that gave tips. "Try the Pomodoro method." "Here's a 5-step framework for your career transition."Users churned after 3-4 sessions. The feedback was consistent: it doesn't really get me.*I went back to basics and studied how actual good coaches work. The pattern:1. **Ask before telling* — a question unlocks more than an answer2. Remember everything — context compounds over sessions; an AI that forgets is worse than a journal3. Call out contradictions — "You said last Tuesday you wanted X, but now you're saying Y. What changed?"## What "remembers context" actually means in practiceThis one took me the longest to get right. When a user comes back after 3 weeks, the coach needs to know:- What they said they'd do- What they actually did (ask them)- What's still unresolved- The emotional tone of previous sessionsBuilding this with LLMs requires careful prompt engineering — you can't just dump the full conversation history in context every time. I use a summarization layer that distills previous sessions into a compact memory object that gets injected at the top of each new conversation.The difference when users experience real continuity: night and day. They stop re-explaining themselves. The coaching can go deeper.## The 4 coaching contextsNot everyone needs the same kind of coaching. We built 4 distinct personas:Personal Growth Coach — habits, identity, who you want to become*Life Coach* — relationships, decisions, big-picture direction*Career Coach* — job search, promotion paths, career pivots*Interview Coach* — specific prep, feedback on answers, confidenceEach has its own approach, language, and depth. Switching between them mid-session is intentional — sometimes a career question is really a life question in disguise.## What we're still figuring outHonestly:- How to make the AI "challenge" users without feeling aggressive- How to handle mental health overlap (we're not therapists — important distinction)- Long-term engagement: how do you coach someone who's achieved their initial goals?## Try itIf you're curious to experience it yourself — or want to understand how the conversation design works — Coach4Life has a free tier (40 messages/month, no credit card).Happy to answer questions about the build if anyone's curious about the technical side — LLM prompt architecture, memory management, persona design.What's been your experience with AI for personal/professional development? I'm curious whether you find the "just ask questions" approach better or if there's a case for more direct AI advice.