Fang Cheng

Fang Cheng is in stealth again. The current bet is called, in the careful language of someone who has done this before, "agentic commerce for restricted wallets" - which is to say, AI agents that can transact on your behalf inside the parts of the financial system where transacting is hardest. Regulated rails. Prepaid balances. The kinds of accounts where a small mistake gets a customer service ticket and a large one gets a lawyer.

If that sentence sounds like a stretch, consider the resume. The pattern is consistent: pick an unsexy operational corner of a giant industry, build software that resolves the problem with less human friction, sell the company. Repeat.

Touchco, her first company, made multi-touch screens that did not require a separate touch layer over the display. Amazon bought it in 2010. The technology ended up inside the Kindle. The press release was tidy. The check, by all reported accounts, was around thirty million dollars. Fang was a few years out of a bioinformatics PhD at NYU at the time, with a paper in Nature already on the CV and a short detour through quantitative high-frequency trading behind her. She has never written the founder-as-Hollywood-myth version of the story, and she has been asked.

After Amazon came Proximiant, a digital-receipts startup nobody quite needed yet, the kind of useful idea that arrives early. Then in 2014 came Linc. Linc was the long one. Linc was where she built the muscle.

What Linc actually did

Linc made customer service software for retail. The product was a conversational layer that sat on top of order systems, returns, shipping, exchanges, and questions about whether the navy chinos run small. It talked over web, email, SMS, Facebook Messenger, voice. It worked with Levi's and PacSun and Vineyard Vines and Carter's. By the time Capacity acquired it in October 2024, Linc was resolving more than eighty-five percent of inbound retail customer inquiries without a human agent. That number is the number she repeats most. It is also the one that mattered to the buyer.

Capacity is an AI platform out of St. Louis with an appetite. The Linc deal was part of a three-company rollup in a single morning - Lucy, Envision, and Linc - paired with a $26 million Series D. Fang stayed on as SVP and General Manager of Retail CX Automation, running the retail vertical for the larger company. Then she left to start something new. The pattern, again.

The part she does not advertise

Early in her career, advisors suggested she change her name. Fiona was floated. Taking her husband's surname, a Jewish one that would have read more comfortably to American investors, was floated. She kept Fang Cheng. Both characters. Original order. The detail is small. It is also the through-line of how she runs companies: shipped without translation, refused to round off the corners.

She has said in interviews that her management philosophy is to treat team members as co-founders, not subordinates. This sounds like a poster on the wall. In practice, at Linc, it meant the Taipei subsidiary office was treated as a peer engineering site rather than an outsourcing back room. It meant the language she uses about the team in podcasts is structurally indistinguishable from the language she uses about herself. There is a kind of leader for whom the word "we" is load-bearing. She is that kind.

From bioinformatics to retail returns

The throughline from a Nature paper on protein interaction to an Alexa skill that processes Levi's exchanges is not as strange as it looks. Both problems are about getting structure out of unstructured signals. A genome and a customer service queue are both noisy. The work, in either case, is figuring out which patterns are real and which are an artifact of how you happened to ask.

This is why she ended up in conversational AI before conversational AI was a fundable category. Linc launched in 2014, six years before the LLM moment that made every retail brand start asking about chatbots. By the time GPT showed up, she had already built the unglamorous infrastructure - the ontology, the workflows, the integrations with order systems that nobody wants to learn the names of - that makes a polite-sounding bot actually solve the problem instead of forwarding it to a tier-two agent. Linc won the AI Breakthrough Award in 2018, when winning that meant something specific.

What she is doing now

The new venture is in stealth, and stealth means she will not name it on a podcast yet. The thesis, as much as has been said publicly, is that AI agents are about to need to spend money on behalf of real users, and the moment they do, the existing payments stack starts to bend. Restricted wallets - the prepaid, regulated, age-gated, expense-categorized corners of the system - bend first. That is where she is building.

If the previous companies are predictive, expect three things. One: a target market most generalists would find boring. Two: a product that gets measured by how many human interactions it removes rather than how many it generates. Three: a small, oddly senior team, run flat, with at least one office that is not in the Bay Area. Whether the third exit follows is a question for around 2028.

What is verifiable now is the resume, and the resume is a long way from common. Three companies. One acquisition by Amazon. One acquisition by Capacity. One stealth venture in flight. A Nature paper. A doctorate in computational biology. A subsidiary office in Taipei that she actually visited. A name she refused to change. A reputation among her former teams as someone who picked up the phone.

The story is mid-stride. She is not waiting for permission to start the next one.