Jolie Ni

I build AI systems for a living and write to find out what I actually think. What follows is an invitation, not a résumé.

  • San Francisco
  • Founder, H.ai

Now

What I am in the middle of.

Running H.ai, an AI consulting and automation firm and an Anthropic Claude partner. Publishing AI in the Field, a newsletter about where AI adoption really breaks. Spending most of my week interviewing PE operators about the parts of their portfolio companies' work AI cannot touch yet.

  • ReadingThe solow paradox in the AI era, what actually moved productivity needle http://digamo.free.fr/david90.pdf
  • Chewing onHow ordinary people rode previous tech bubbles, and what they saw too late.

Open questions

Six things I would rather talk about than the weather.

  1. Why adoption stalls

    The blocker is almost never model capability. It is that nobody has settled who captures the gain: the model company, the enterprise, the manager, or the person actually doing the work. Misattributed profit kills more rollouts than bad accuracy does.

  2. Every bubble, from the ground floor

    I read the history of tech bubbles with one question each time. What did an ordinary person actually do to ride the wave, and what did they understand only too late?

Made

Things that exist because I made them.

  • H.ai 2026 →

    AI consulting and automation for teams that have run out of patience with pilots. Anthropic Claude partner.

    I was laid off and decided not to job hunt. The bet was that the interesting problem in AI right now is not capability, it is getting capability to survive contact with a real company. So I go into businesses that have tried AI and been disappointed, find the place where the work actually gets stuck, and build the narrow thing that unsticks it. Mostly that means understanding an industry well enough to know which step nobody wants to do.

  • Counterfactual poker 2022

    Open-source Texas Hold'em AI using Counterfactual Regret Minimization, trainable on a personal computer.

    I brought the research topic to the professor rather than picking from his list, then led a team of three engineers. It won the UofT Sentinel Senior Project Award and the ECE Capstone Distinction Award.

    The part I still care about is not the poker. It is the constraint. We were asking how much competence you can extract from a hard compute ceiling, which is the same question that decides whether access to AI ends up concentrated in the few places that can afford the compute. I picked the problem before I had the language for why it mattered.

  • Flexible Workforce Payment Architecture Patent

    US patent, filed while building workforce management and banking infrastructure at Square.

    Payroll and scheduling infrastructure at a scale where a rounding error is somebody's rent. I also pitched a hackathon project to Jack Dorsey, which taught me more about how to frame an idea in ninety seconds than any amount of documentation ever did.

Route

Four cities, and what each one took.

  1. Beijing

    until 2014

    Hometown with a lot of childhood memories

  2. Vancouver

    2014 – 2017

    High school, met so many chinese international school with similar background

  3. Toronto

    2017 – 2025

    Engineering at UofT. Tears and hardwork. SDE at Square, made good friends

  4. San Francisco

    2025 – 2026

    Met so many exceptional people, truly eye opening

Writing

I write in two languages and argue with myself in both.

  • AI in the FieldNewsletter. Where adoption breaks, reported from inside real companies.
  • 朱莉奇遇记Xiaohongshu, in Chinese. Founding, money, privilege, two cultures.

Talk to me about: AI adoption, Feminism, Chinese Culture, Philosiphy

If one of those questions is also yours, write to me.

I would rather hear from someone who disagrees with the fifth one than from someone who liked the layout. No pitch necessary. Tell me what you are working on and where you are stuck.

jolieni@hconsult.ai