ChemMind Technologies founder Ning Xia has spent 18 years building AI tools for pharmaceutical chemical synthesis, and he has a clear answer to whether today's AI matches the capabilities of a chemist with a decade of experience: for primary intellectual work, yes - and that changes what chemists should focus on.
Xia, who holds a doctorate in organic chemistry, sat down for a conversation that covered the founding of ChemMind, the difference between black-box and white-box AI systems, and the importance of customer feedback loops that investors often overlook.
He started in 2008 in France, building what later became ChemMind Technologies, during a period when the field had no data, no investors, and no off-the-shelf AI infrastructure. "We relied entirely on personal ideals and interests," Xia said. "Moreover, there were no underlying AI technologies as there are now, no data, and all infrastructure had to be built from scratch."
From individual contest win to million-dollar contracts
In 2016, while working at a chemical B2B e-commerce company, the company registered the chemical.ai domain and built a retrosynthesis prediction system. Xia decided to show it to WuXi AppTec executives - without a company to represent.
"We had contacts with some executives of WuXi AppTec, so I said that I had such a set of software, would you like to try it?" he said. "I demonstrated on the spot, ran various molecules, and the effect was very good, so we had the first commercial partnership with WuXi AppTec."
That same year, he co-founded ChemMind Technologies with a team of just a dozen or so people. The first major milestone came when WuXi AppTec introduced ChemMind to one of its large pharmaceutical customers, which contracted the company for an AI retrosynthesis module in a deal worth more than 1 million USD. The customer was building an AI pharma system - with the retrosynthesis war completely in their hands. That was an early test of how well a research tool used in the industrial closed loop.
"When your product first comes out, it's definitely not 90 points, maybe 60 or 70," Xia said. "We actually experienced a lot of customer feedback, iteration and optimization." The company kept a demand list - "there may be thousands of opinions put forward by various customers, each of which means a lot of work," he said.
White-box vs. black-box AI
A recurring theme in ChemMind's approach is its preference for white-box AI, which explicitly uses known chemistry rules and formulas, over pure black-box that takes data in and produces results without explanation. Xia explains it's a matter of trust.
"For such serious scientific software, if it is a black-box, it is difficult to build customer trust," he said. "It's non-interpretable. The biggest challenge is that you find a problem but don't know where the problem is, so you have no way to improve it."
ChemMind's system uses black-box models for their predictive power, but all their conclusions are checked, explained, and constrained by the white-box model.
"All conclusions given by large models or black-box models must be explained, verified and constrained in our white-box model, so that we can eliminate hallucinations," Xia said, and this combination is what lets researchers know which of the model's ambiguous conclusions is right and which is wrong.
He also pushed back on the notion that great technology immediately produces the best product. "There's a lot of work that is needed in between, which many people, whether investors or the industry, tend to ignore - that in many cases, the payoff comes through iteration, not in one big success."
From graduate students to pharmaceutical scale-up
The tool has different value at each points along a drug's development chain. Molecular designers use it to check whether a molecule can be synthesized. Medicinal chemists use it to quickly plan routes when they need a few milligrams of a compound. Process chemists, however, have opposite demands: they only need to make one molecule but they need to do it at the lowest possible cost, even if it means using a riskier reaction.
Asked what this means for students, who traditionally spent months exploring synthetic routes by hand under a supervisor's guidance, Xia is direct: "Probably all the work of students can now be done with the help of AI. The first thing AI will replace is probably this kind of primary intellectual work."
Unlike programmers, however, chemists may have more room to move up. "The field is more optimistic," he said. "We have too many innovative materials to develop, and the space is large enough."
He also responded to a question about whether he would would have sold the company during the 2020-2021 biotech boom, when a large CRO sent acquisition offers multiple times. "Recalling back today, I don't believe selling would have been a beneficial outcome," he said. "I'm a relatively idealistic entrepreneur."
Why this matters for people in research and development
For working scientists in R&D settings, the takeaway is twofold. First, AI tools are being used to handle more and more of the repetitive, literature-heavy, trial-and-error work that has traditionally been the starting point for junior chemists. That shift is real and already happening - as Xia puts it about graduate students: "The first thing AI replaces is abundance of primary intellectual work." Everyone who works in a lab must be prepared to add value beyond that level, the kind of creative problem-solving for which there is no template.
Second, the lesson on iteration matters beyond this tech vendor. Xia's experience demonstrates that a scientific tool's real usefulness depends heavily on customer-driven adaptation: thousands of feedback items, dozens of refinements, and a willingness to let white-box logic constrain black-box findings so the system can be audited and improved. For any lab or research organization selecting an AI tool, that means it's worth asking not just about model performance on benchmarks, but about how the vendor responds to real-world feedback and how transparent the system is about its own predictions and errors.
Your membership also unlocks:
