A Developer Used AI to Find a Planet in NASA's Old Data

The signal is tiny. Every 3.18 days, a star roughly 116 light-years away dims by about 0.05 percent for two hours. That is the kind of data point that usually gets buried in the noise of a massive telescope archive. It sits there, unremarkable, until someone specifically looks for it. Pavel Rabtsevich looked. He is not an astronomer by training, but he is a developer who spent two weeks running 74 separate analyses and writing over 1,000 scripts to verify a pattern he found in NASA's public records.

This is not a story about a massive observatory or a government budget. It is a story about a single individual using AI coding tools to act as a space detective. Rabtsevich used Anthropic's Claude Code and OpenAI's Codex to process observations from the Transiting Exoplanet Survey Satellite, or TESS. He did not just find a blip. He traced the signal back through datasets from 2018 and 2020 to confirm consistency. The result is a candidate planet about 1.4 times the size of Earth, and NASA has now scheduled follow-up observations to test the claim.

From Code Assistants to Scientific Tools

Most people think of AI coding agents as tools for writing software. They help you debug, refactor, and automate repetitive tasks. That is their primary job. But Rabtsevich applied that same logic to a completely different problem: processing telescope observations. He needed to fit potential planetary transits and check alternative explanations for the dimming star. The AI handled the heavy lifting of data processing and script generation, allowing him to focus on the scientific interpretation.

This shift is significant. It shows that these tools are no longer just text generators. They are becoming capable of handling complex, multi-step analytical workflows. Rabtsevich used the agents to run dozens of tests, including checks for nearby stars and instrumental glitches. These are the exact kinds of errors that can mimic a planetary transit. By automating these checks, he could rule out false positives with a speed and consistency that would be difficult to achieve manually.

The Challenge of Verification

Finding a pattern in old telescope data is one thing. Establishing that it represents a real planet is another. This is where the scientific method kicks in. Rabtsevich knew that a simple dip in brightness could be caused by many things other than a planet. It could be a star spot, a binary star system, or a sensor artifact. The AI agents helped him navigate this complexity by systematically testing each hypothesis.

The candidate remains unconfirmed as of now. That is the honest reality of the situation. However, the fact that NASA has scheduled new observations based on his findings is a major validation. It means that the data is robust enough to warrant further scrutiny by professional astronomers. This is a sharp shift in how individuals can interact with deep space archives. The barrier to entry for scientific discovery is lowering, not because the science is getting easier, but because the tools for handling the data are getting smarter.

A New Era of Citizen Science

This case illustrates how AI agents are democratizing scientific discovery. In the past, analyzing TESS data required specialized expertise and access to high-performance computing resources. Now, a developer with a laptop and a subscription to an AI coding tool can perform similar analyses. Rabtsevich’s work was not a fluke. It was a structured investigation that leveraged the ability of AI to handle repetitive, data-heavy tasks.

The implications go beyond this single discovery. If a developer can find a hidden planet, what else is hiding in the public data? The TESS archive contains millions of observations. Most of them have been scanned by automated algorithms, but those algorithms are not perfect. They can miss subtle signals or misclassify them. Human curiosity, guided by AI tools, can pick up where the automation leaves off.

The Road Ahead

We will know more once the follow-up observations are complete. If the signal holds up, Rabtsevich will have made a genuine contribution to our understanding of the universe. If it does not, he will have demonstrated the power of new tools to explore the unknown. Either way, the story is about the changing nature of discovery.

It is a reminder that the frontier of science is not just about building bigger telescopes. It is also about finding better ways to use the data we already have. The next big discovery might not come from a new observatory. It might come from a developer, a weekend, and a few thousand lines of AI-generated code.