How AI Is Making Scientific Tools Easier to Use

A company called SandboxAQ has found a way to let researchers use powerful molecular modeling tools through conversational AI—just by talking to Claude, an AI assistant made by Anthropic. Instead of requiring specialized software that takes years to learn, scientists can now ask questions in plain language and get answers about how drugs might work or how materials might behave.

This connection works through something called the Model Context Protocol, which acts as a bridge between Claude and SandboxAQ's specialized computational tools. Think of it like translating between two languages: Claude speaks in natural conversation, while the scientific tools speak in numbers and physics. The protocol lets them talk to each other seamlessly.

Making Hard Science Accessible

SandboxAQ's tools can perform calculations that regular AI systems cannot do on their own. They can predict how molecules will bind together, model chemical reactions, and simulate how materials will behave under different conditions. These are the kinds of calculations that once required scientists to use expensive, difficult-to-learn software packages.

Two of the company's most useful tools are now available through Claude conversations. One predicts whether a potential drug molecule will actually stick to its target in the body. The other simulates how materials absorb chemicals—useful for everything from water filtration to battery design. Researchers can get answers without switching between multiple programs or writing code.

On a recent project hunting for new treatments for brain diseases, SandboxAQ's approach expanded the search space from 250,000 possible molecules to 5.6 million, and found candidate drugs 30 times more often than traditional methods. The company ran these calculations using Google Cloud's computing infrastructure.

Real-World Partnerships

SandboxAQ is working with established pharmaceutical companies and investment funds to expand who can use these tools. One partnership with a major biopharma company focuses on nervous system drugs and includes potential payments reaching $200 million as the work progresses. A sovereign investment fund in Bahrain is also partnering with SandboxAQ to build biotech research capabilities in the region.

Anthropic, the company behind Claude, simultaneously launched Claude for Life Sciences, adding connections to tools that scientists already use regularly—lab data platforms, scientific illustration software, research databases, and genomics data. This suggests both companies are working together to make Claude a central hub for scientific research.

What This Means

The bigger story here is something I have watched happen before in technology. Thirty years ago, when the internet went commercial, people needed to understand servers and networking to run a website. Then web hosting came along and hid that complexity behind simple interfaces. The same thing happened with cloud computing—managing servers became pushing buttons on a dashboard. Now we are seeing it again with scientific computing: the complexity of molecular modeling is disappearing behind a conversational interface.

If this approach works at scale, it could change who gets to use these tools. Right now, only researchers with specialized training in computational chemistry can run these simulations easily. If a biologist or materials engineer can get the same insights just by asking Claude, that opens up the work to many more people and institutions.

How It Actually Works

When a researcher asks Claude a question about molecules or materials, the AI doesn't do the calculation itself. Instead, it translates the question into a format that SandboxAQ's tools understand, runs the specialized computation, and brings the results back into the conversation so the researcher can see both the numbers and what they mean in plain language.

This design keeps the heavy mathematical work separate from the conversation interface. It means the tools can stay specialized and accurate while still being easy for anyone to use. SandboxAQ's team of 70 people in the drug discovery space makes sure that the suggestions the system offers align with how real researchers actually work.

The Open Question

Whether this approach works long-term depends on whether talking to an AI is actually good enough for the messier parts of scientific work. Real research involves asking follow-up questions, tweaking parameters, and trying different approaches. Early signs suggest that researchers who already rely on Claude for literature review and planning are the ones most likely to expand into using it for calculations too. But how far this extends into daily research practice remains to be seen.