A tech-industry product manager who isn’t an astronomer used Claude Code to find a possible undiscovered exoplanet.
His name is Pavel Rabtsevich.
He used Claude Code and Codex to analyze publicly available observations from NASA’s TESS space telescope.
Around the star TIC 4206066, about 116 light-years from Earth, he spotted something unusual:
Every 3.18 days or so, the star dims by about 0.05% for roughly two hours.
This could mean a planet regularly passes in front of the star.
To rule out false positives, he spent about two weeks conducting 74 analyses, generating over 1,000 code scripts, and comparing independent observations from 2018, 2020, and 2025.
Most interestingly, he deliberately withheld the data from one year and used the other two years’ data to predict when the planet would transit.
All three tests matched the predictions.
The candidate planet is currently estimated to have a radius about 1.4 times that of Earth.
He also successfully applied for follow-up observations with TESS, which are scheduled to take place again from October 31 to November 26.
However, it remains an unconfirmed planet candidate and will require further observations and scientific validation.
What’s truly noteworthy about this experiment is that AI coding agents aren’t just for building websites—they can also help individual researchers create end-to-end scientific data analysis workflows.
Original source:
https://x.com/p_rabtsevich/status/2107803696058822742
3D interactive project:
https://tic4206066.pages.dev
Data/image sources: Pavel Rabtsevich / NASA TESS
His name is Pavel Rabtsevich.
He used Claude Code and Codex to analyze publicly available observations from NASA’s TESS space telescope.
Around the star TIC 4206066, about 116 light-years from Earth, he spotted something unusual:
Every 3.18 days or so, the star dims by about 0.05% for roughly two hours.
This could mean a planet regularly passes in front of the star.
To rule out false positives, he spent about two weeks conducting 74 analyses, generating over 1,000 code scripts, and comparing independent observations from 2018, 2020, and 2025.
Most interestingly, he deliberately withheld the data from one year and used the other two years’ data to predict when the planet would transit.
All three tests matched the predictions.
The candidate planet is currently estimated to have a radius about 1.4 times that of Earth.
He also successfully applied for follow-up observations with TESS, which are scheduled to take place again from October 31 to November 26.
However, it remains an unconfirmed planet candidate and will require further observations and scientific validation.
What’s truly noteworthy about this experiment is that AI coding agents aren’t just for building websites—they can also help individual researchers create end-to-end scientific data analysis workflows.
Original source:
https://x.com/p_rabtsevich/status/2107803696058822742
3D interactive project:
https://tic4206066.pages.dev
Data/image sources: Pavel Rabtsevich / NASA TESS