Over the past few years, AI has made remarkable strides in the digital realm. From molecular design and drug discovery to the generation of scientific hypotheses, AI is continually expanding the boundaries of human exploration into the unknown. But in scientific research, coming up with an idea is often just the first step.
Can a molecule actually be synthesized? Do its properties match expectations? Can experimental results be reliably reproduced? These questions ultimately come back to the lab, where answers are found through real, hands-on operations and validation.
In a sense, the better AI becomes at computation, the more the bottleneck shifts to experimental execution.
Take drug discovery as an example. Scientists might spend only a fraction of their time designing reaction pathways. What truly consumes enormous effort is the subsequent cycle of repeated experiments, sample processing, and purification. Even when a model has produced an answer worth validating, there's still a mountain of work in the lab that must be done by human hands.
So what if AI could do more than just think and decide? What if it could step into the lab and personally carry out the experiments?
This is the question HeTan Intelligence has been exploring.
This past March, HeTan Intelligence's Physical AI Scientist, Talos, operated continuously for 50 days in a real lab at a leading pharmaceutical company, completing over 8,000 physical operations with zero errors recorded. Behind these numbers lies an important step in AI's journey from the digital world into real scientific settings.
Recently, HeTan Intelligence also signed a strategic agreement and purchase order with global life sciences company Danaher to jointly advance an AI-robot autonomous bioprocess development laboratory. This marks Danaher's first embodied intelligence BEACON project in China.
In this episode of Yunqi Capital's brand podcast "Attent!on," we invited Zhigang Chen, founder of HeTan Intelligence, Zhuo Fang, Chief Science Officer of Danaher China, and Yu Sang, investor and executive director at Yunqi's frontier technology team, for a conversation.
From Chen's career across internet, healthcare, and AI, to how Talos completes a real molecular purification experiment; from the limits of traditional lab automation, to how Danaher and HeTan Intelligence are jointly exploring next-generation autonomous laboratories — this dialogue attempts to answer a more fundamental question:
When AI can already generate more and more scientific hypotheses, who will turn them into real-world answers?
You'll hear about:
From digital intelligence to physical execution: Why can AI already design molecules, yet experimental validation remains a critical bottleneck in the R&D chain?
The boundaries of lab automation: Why does traditional automation struggle to cover complete experimental workflows? What long-standing problems can robots solve?
Deconstructing Physical AI Scientist Talos: From sample identification and experimental design to purification and detection, how does AI complete a full molecular purification?
From demo to real-world delivery: Behind 50 days, 8,000+ operations, and zero errors, how does the robot handle complex environments and experimental demands?
The real standards of industry partnership: Why did Danaher choose HeTan Intelligence? In lab intelligence, is speed, stability, or reproducibility what matters most?
The next step for AI for Science: When experimental data flows back into models, can an intelligent closed loop form from molecular design to physical validation?
This episode's guests
Zhigang Chen Founder of HeTan Intelligence; former Chief Architect at Alibaba Health Information Technology Limited, founder of Tencent's medical big data lab, and first Chief Digital Officer at WuXi AppTec.
Zhuo Fang Chief Science Officer, Danaher China
Yu Sang Investor and Executive Director, Yunqi Capital Frontier Technology Team
· Episode Timeline
Starting point: Why choose to bring AI into the lab?
00:01 Opening: 50 days, 8,000+ physical operations, zero errors — how does a robot become a Physical AI Scientist?
03:06 From internet, healthcare to AI: Chen's entrepreneurial journey and Yunqi's early investment thesis
07:03 From automation to robotics: The technological inflection point brought by VLM and VLA, and why wet labs are critical infrastructure for AI
From traditional automation to Physical AI Scientist
09:07 The limits of lab automation: multi-equipment coordination, sample transfer, and consistency of experimental results
13:20 How does Talos complete a small-molecule purification? From sample identification, TLC spotting, to automated column chromatography and detection
17:52 How do robots adapt to complex lab environments? Structured operations, model training, and Sim-to-Real
21:04 Next-generation intelligent labs: How do robots, automated workstations, and mobile equipment collaborate?
From demo to real-world delivery: The test of industry partnership
24:05 From AI synthesis models to Physical AI Scientist: How did the entrepreneurial direction gradually clarify?
27:07 Real experimental results equivalent to experienced scientists: How far are robots from stable delivery?
29:16 Why did Danaher choose HeTan Intelligence? From BeiGene to China's first embodied intelligence BEACON project
33:02 How do industry players evaluate intelligent labs? Speed, throughput, stability, and consistency
34:55 Entering real labs: From biosafety cabinet disinfection to scalable replication across different customer scenarios
39:16 HeTan Intelligence's core moat: cross-disciplinary cognition, R&D delivery system, and AI-native operations
AI for Science's next stop: Autonomous labs and data closed loops
40:57 Industry progress in AI for Science: How far has molecular design come?
42:49 Next-generation autonomous labs: How do AI, agents, and robots collaborate? How does experimental data feed back into models?
45:28 HeTan Intelligence's future exploration: Making experimental validation form faster closed loops
The scientist's next lab partner
47:12 If robots take over repetitive experiments, how will scientists' work change?
48:48 From physical-world validation of scientific hypotheses: Giving humanity better tools to explore the unknown
50:22 Yunqi investment perspective: Why continue to focus on AI for Science?
· Terms that may need brief explanation
AI for Science (AI4S): The direction combining artificial intelligence with scientific research, using AI to assist molecular design, scientific computing, experimental planning, and scientific discovery.
Physical AI: Bringing AI's perception, decision-making, and execution capabilities into real physical environments, enabling AI to complete real-world tasks through robots and other carriers.
Wet Lab: Experiments conducted in real labs using reagents, samples, and instruments, as opposed to computer simulation and theoretical calculation (dry labs).
Dry Lab: Research primarily relying on computers, mathematical models, and data analysis, without requiring direct hands-on lab operations. Examples include molecular structure prediction, drug molecule design, computational simulation, and scientific data analysis.
Talos: HeTan Intelligence's Physical AI Scientist product, combining an AI molecular brain with Physical AI execution capabilities to explore intelligent execution of complex laboratory tasks.
Small-molecule purification: The process of separating target products from unreacted materials, byproducts, and other mixtures after a chemical reaction is complete — an important experimental step in drug discovery.
TLC (Thin Layer Chromatography): A common chemical analysis method that helps judge sample composition and purification conditions by observing how different substances separate on a thin-layer plate.
LC-MS (Liquid Chromatography-Mass Spectrometry): A detection technique combining liquid chromatography separation with mass spectrometry analysis, used to analyze compounds and related information in samples.
Sim-to-Real: The technical transfer process from simulated to real physical environments, where robots complete training and validation in simulation before executing tasks in real scenarios.
VLM (Vision-Language Model): Models capable of simultaneously understanding visual information such as images and video together with natural language.
VLA (Vision-Language-Action Model): Combining visual perception, language instructions, and action execution to enable robots to take action based on environment and task instructions.
Workcell: A relatively independent work area in lab automation, typically containing specific instruments and corresponding operational procedures.
BEACON project: Danaher's first embodied intelligence project in China mentioned in this episode, jointly advanced by HeTan Intelligence and Danaher to develop an AI-robot autonomous bioprocess development laboratory.
🏢 About Yunqi
Founded in 2014, Yunqi Capital is an early-stage lead investor focused on digital intelligence and hard technology. We have been repeatedly named among the "China Best Early-Stage Investment Firms TOP 10" by Zero2IPO, ChinaVenture, 36Kr, and other institutions.
Over the past 12 years, we have accompanied more than 200 technology startups in their growth, including industry leaders such as MiniMax, JD.com, Manycore Tech, DeepRoute.ai, Neolix, Keenon, Independent Variable Robotics, Astribot, RealMan, Noematrix, PingCAP, Zilliz, ASTRONSTONE, and XTransfer.
📩 Contact Us
For pitch decks or ecosystem partnership: community@yunqi.vc
WeChat Official Account: Yunqi Partners (ID: yunqipartners)
Xiaohongshu: Yunqi on Anfu Road
View episode transcript on Xiaoyuzhou