Tetsuwan Scientific is trying to solve a practical problem in modern labs: too much scientific work still depends on slow, repetitive manual handling. Its answer is not a humanoid machine, but a glass-box lab robot connected to AI systems that can evaluate results, adjust experiments, and move closer to acting like an independent scientific operator.
From lab frustration to a robotics startup
The company began after Cristian Ponce met Théo Schäfer at a Halloween party in 2023 hosted by Entrepreneur First. Ponce was dressed as Indiana Jones. Schäfer brought a background that included a master’s in underwater autonomous robots at MIT and work at NASA’s Jet Propulsion Lab exploring Jupiter’s moons for alien life.
Ponce had been at Cal Tech doing bioengineering work with E. coli. The two found common ground in a less glamorous part of science: the day-to-day burden of lab technician work. Genetic engineering, in particular, can require long stretches of moving liquids by hand with a scientific syringe known as a pipette.
That manual labor matters because experiments often need small adjustments. In many labs, a human can still be the easiest, cheapest, and most precise option because existing robotic systems tend to be specialized, expensive, and dependent on programmers. When researchers change an experiment’s parameters, the robot may need new code, debugging, and setup before it can continue.
Tetsuwan Scientific first aimed to improve that situation by modifying lower-cost white label lab robots. The goal was not simply to replace one manual step, but to make robotic automation practical when experiments keep changing.
The AI moment that changed the plan
In May 2024, the co-founders were watching OpenAI’s multi-model product launch, the one connected in the source account to a sound-alike voice that ticked off Scarlett Johansson. What stood out to them was not only the model’s conversational interface, but what it suggested for scientific reasoning.
Ponce then used GPT-4 with an image of a DNA gel. According to the source article, the model recognized what the image showed, identified an unintended DNA fragment called a “primer dimer,” and gave a detailed scientific suggestion about what might have caused it and how conditions could be changed to prevent it.
For Ponce, that exposed the missing connection between AI analysis and laboratory action. He described it as a “light bulb moment,” because large language models could already help diagnose scientific outputs, but had “no physical agency to actually perform the suggestions that they’re making.”
That distinction is central to Tetsuwan Scientific’s thesis. A model that can interpret a result is useful, but a robot that can also act on the next step could change how experiments are run. The company’s work sits in that gap between digital recommendation and physical execution.
Why scientific intent is hard to automate
The source article makes clear that Tetsuwan Scientific is not the first group to explore robotic AI scientists. The idea can be traced back to 1999 with Ross King’s robots “Adam” and “Eve,” and the field gained momentum through academic papers starting in 2023.
Still, Tetsuwan Scientific’s research found a specific missing layer: software that can translate scientific intent into robotic execution. In plain terms, a scientist may know what an experiment is trying to test, but a robot needs that intent converted into specific actions it can perform reliably.
That translation is not as simple as telling a machine to pipette a liquid. The robot also needs context about the physical materials it handles. Ponce gave examples: a liquid may be viscous, or it may crystallize. Those qualities affect how a lab robot should act, but they are difficult to encode with rigid instructions alone.
Audio LLMs, with hallucinations reduced by RAG, are part of the company’s approach for handling details that are hard to code directly. Tetsuwan Scientific is also building software and sensors so its robots can understand calibration, liquid class characterization, and other properties.
- Scientific intent: what the experiment is meant to investigate.
- Robotic execution: the physical steps the machine must perform.
- Context: material properties such as viscosity or crystallization risk.
- Feedback: evaluating results and changing the next action.
Early customers and the larger ambition
Tetsuwan Scientific currently has an alpha customer: La Jolla Labs, a biotech company working on RNA therapeutic drugs. The robots are being used to help measure and determine dosage effectiveness.
The startup has also raised $2.7 million in an oversubscribed pre-seed round led by 2048 Ventures. Carbon Silicon Ventures, Everywhere Ventures, and influential biotech angel investors also participated.
The near-term product is a robotic system that can support lab experimentation with more autonomy than conventional automation tools. The long-term ambition is larger: independent AI scientists that can automate the scientific method from hypothesis through repeatable results.
Ponce framed that destination in unusually sweeping terms. “It is the craziest thing that we could possibly work on. Any technology that automates the scientific method, it is the catalyst to hyperbolic growth,” he says.
Tetsuwan Scientific is not alone in that belief. The source article also names FutureHouse and Seattle-based Potato AI as others working on AI scientists. The competition suggests that autonomous science is moving from a research concept toward a startup category, even if the hard work remains in making robots understand enough of the lab to act safely, precisely, and usefully.