What if you could turn dozens of scientific research papers into an interactive laboratory you can actually experiment with?
Not just summaries. Not another chatbot explaining complicated equations.
Actual working simulations built directly from published research.
A fascinating Claude Code experiment attributed to Boris Cherny, the creator of Claude Code, shows what happens when you give 12 AI agents 64 scientific papers about fluid dynamics and ask them to build a browser-based laboratory.
And the result is beautiful.
Here's what the agents reportedly created:
Interactive fluid simulations: Choose different experiments and watch physical phenomena unfold.
Adjustable parameters: Change conditions using sliders and immediately see how the results respond.
Multiple simulation methods: Compare different approaches using the same experimental settings.
Scientific references: Explore equations, source code, and measurements alongside each animation.
Research verification: Compare simulation outputs with the results reported in the original papers.
But here's the really interesting part.
The agents didn't all work on the same problem.
They had specialized responsibilities.
Research agents extracted equations, assumptions, and experimental parameters from the papers.
Development agents translated those methods into working simulation code.
Interface agents built the interactive controls and visualizations.
Review agents checked the implementations against the original research, looking for numerical errors, instability, and discrepancies.
According to the circulated demonstration, the project covered 11 simulation methods across 28 reference scenarios, with seven meeting the selected validation tolerances and four requiring further review.
Those figures are reported rather than independently verified.
And that distinction matters.
A beautiful simulation isn't necessarily a scientifically accurate simulation.
But imagine applying this workflow to your own research.
A collection of scientific papers becomes an interactive environment where you can test assumptions, compare methods, adjust variables, and inspect the underlying code.
Want to try something similar?
Here's a prompt you can adapt for Claude Code:
"Turn these research papers into an interactive browser-based laboratory.
Extract the methods, equations, assumptions, and experimental parameters from each paper, including page references.
Assign different methods to specialized agents, using shared inputs and measurement tools so the results can be compared.
Have independent review agents verify each implementation against the original paper. Document discrepancies and clearly mark unverified results.
Build an interactive interface with parameter sliders, visual simulations, method comparisons, and access to the source code behind each experiment.
Make the experiments reproducible, including their dependencies, reference data, and validation procedures."
Start with one paper and one experiment, verify the results, and expand from there.
And here's what I find genuinely exciting.
We're moving beyond AI that simply reads research papers and tells us what they say.
We're approaching a world where AI agents can help turn scientific knowledge into working, explorable software.
From reading science to experimenting with it.
That's a much more interesting use of AI than generating another thousand summaries nobody will ever read.
#ClaudeCode #AIAgents #ScientificResearch #FluidSimulation #CreativeCoding #AIResearch
YouTube / ABVX