Jose Zea

Every healthcare AI startup brags about accuracy. 95%. 97%. 99%. The ones that win aren't the most accurate — they're the ones a doctor can't imagine working without. Don't be the best. Be the only.

Being "nice" but vague is not kindness. It's cowardice. Most leaders avoid hard feedback, then wonder why performance drops. Clear without care = fear. Care without clarity = chaos. The formula: high standards, low ego, own mistakes fast.

Most people in meetings aren't listening. They're loading their next sentence. Real listening: no defending, no rehearsing, no fixing. Just absorb. The founder who listens longest usually decides best.

Two people. Same AI tools. Different outcomes. One used AI to do his job faster — auto-generate reports, draft emails, summarize meetings. He got efficient at tasks a bot now does for free. One used AI to do work she couldn't do before — found a pattern in 3 months of data no o

Growth doesn't fix character. It exposes it. When you get leverage, everything gets louder — your strengths, your blind spots, your ego. Before another growth playbook, build a character framework. Scaling is hard. Scaling without losing yourself is harder.

Next time you’re “one more meeting” away from deciding — stop. Ship the rough version. Watch real users react. Learn what no plan could tell you. Thinking gives you theories. Shipping gives you evidence.

A product question I use in reviews: Telegraph or telephone? If users must learn your language, adoption stalls. If your product speaks theirs, adoption compounds. Great UX feels familiar on day one.

How the highest-output people actually use AI: 9:00 — Write a one-paragraph brief 9:05 — Three AI agents start on different parts 9:20 — Review. Catch wrong assumptions. Redirect. 9:40 — Merge. Rewrite the one part only a human gets right. 10:00 — Ship a two-day task in one hour

How Dr. Javier Flores uses Claude to treat patients in Ecuador: Step 1 — He diagnoses. Sees the patient. Examines. Decides. (AI touches nothing here.) Step 2 — Uploads labs + his clinical framework + latest guidelines to Claude. Step 3 — Asks: "Given these exact values, flag any

I used to call many constraints “reality.” Now I call many of them “habits.” Builders ask a different question: not “how does this system work?” but “what should this system become?” The world is more editable than it looks.

Anyone can copy your feature list. Almost no one can copy your taste. Taste = the compounding effect of a thousand decisions: what you cut, what you keep simple, what you refuse to ship. That's a moat no competitor can reverse-engineer.

Your friend paid $2,000 for an AI automation course. Last week, Claude did the same thing in 4 minutes. AI skills expire faster than you can learn them. What doesn't expire: being the person people call at 2am when the system breaks and there's no playbook. Stop collecting AI

The bottleneck in AI isn't execution speed. It's imagination survival. We're so good at shipping fast that we crush ideas before they can breathe. New ideas need a maternity ward before they need the arena.

We had 48 hours before a product launch. Not enough time to fix the part of the UI that felt "weird." So we shipped it weird. That "weird" part became the feature users talked about most. More time would've killed it. We would've sanded it down until it looked like everything

Nobody warned me about this part of my thirties. Guillain-Barré paralyzed my legs and arms. I went from running a company to relearning how to grip a fork. When you can't do anything, the only question left is: who am I without the work? Things don't slow down. Sometimes they

"What do you think?" → a list of everything wrong. "What would you do if you were building this?" → a co-builder. One word changes the entire room.

Fun Claude Code command: /cost-estimate Scan your codebase and calculates what's the cost of your codebase' Our codebase at arkangel.ai: 8,900 hours of work. Human cost: $3,115,000. Claude cost: $17,800.

How AI reads a medical chart (step by step): 1. Ingests the full record — digital, scanned, even handwritten 2. Identifies every clinical condition mentioned 3. Maps each one to its ICD-10 code 4. Links every code to the exact sentence in the document 5. Outputs an annotated PDF

300 million people live with a rare disease. The average diagnosis takes 5+ years. A new Nature paper introduces DeepRare — an AI agent that outperformed experienced physicians at diagnosing rare diseases (64.4% vs 54.6%) with traceable reasoning. Not a black box. A second brai

AI is still in first-date mode☕️: great conversation, low trust, no shared routine yet. That is why demos look amazing and production feels messy. The winners won’t be the loudest launches. They’ll be the teams that make AI useful on a normal Tuesday.