A star 116 light-years away dims by a hair every 3.18 days. Pavel Rabtsevich, a 28-year-old product manager in Spain with no astronomy training and no telescope, says he spotted that pattern using AI coding agents and public NASA data. In a few weeks, a space telescope will check his work.
That’s the part that makes this worth your attention. It isn’t a vague claim that a chatbot “discovered a planet.” It’s a specific, falsifiable prediction with dates attached, and he has posted those dates publicly.
“If the star dims on time in November, this gets a lot more serious,” Rabtsevich wrote in an X post outlining his findings. “If it doesn’t, I’ll post that too.”
What he found, and where
The star is TIC 4206066, a little smaller than our Sun. The signal is a slight, regular drop in its brightness every 3.18 Earth days, the kind of dip you’d expect if something were crossing in front of it.
Rabtsevich calculated that the possible planet would be just a little over 1.4 times bigger than Earth. With a year that lasts just over three Earth days, he predicts a surface temperature around 1,000 degrees Fahrenheit. That’s significantly hotter than Mercury and hot enough to melt lead.
There’s a second, weaker signal too. His AI-assisted digging turned up another dimming pattern on the same star, once every 11.13 Earth days, which could point to a second previously unknown exoplanet. Rabtsevich considers the evidence for that one less convincing, and it’s good that he says so up front.
Three AI agents, one human making the calls
The raw material came from the Transiting Exoplanet Survey Satellite (TESS), a space telescope operated by MIT astronomers and launched by NASA in 2018. TESS hunts for planets outside our solar system by tracking stellar brightness. If a star dims at regular intervals, a planet may be passing in front of it and blocking some of its light.
Rabtsevich worked from a narrow slice of that data, captured from November of last year to early January. He said the dataset covered more than 126,000 stars. He fed it to TypeSafe’s Jev, Anthropic's Claude Code and OpenAI’s Codex, prompting them to look for dimming patterns and narrowing the candidate list round by round.
He was clear that the models didn’t run the show.
“I didn’t take the models’ first answers,” Rabtsevich said. “I pushed them, round after round, to improve their own methods, and checked what they gave me against other models. “I decided which tests to run, judged whether a result was strong enough, and chose what to keep.”
That’s the workflow I’d trust, if any. Cross-checking one model against another is the cheapest defense against a confident wrong answer, and it’s the step most people skip.
74 attempts to prove himself wrong
Once TIC 4206066 surfaced, Rabtsevich went back further. He pulled the star’s TESS measurements from 2020 and 2018. The same dips showed up in both years, at the same frequency.
He then ran a total of 74 tests with Claude Code, all aimed at disproving his own hypothesis. In one, he gave the agent two of the TESS datasets and asked it to predict when the dimming would occur in the third. Its predictions were correct in all three cases.
Rabtsevich sent that report to MIT and asked that TESS observe TIC 4206066 the next time the telescope points at that sector of sky. The request was approved. From October 31 to November 26, the telescope will measure the star’s brightness every two minutes.
Earlier this week, he publicly posted the times when the star should dim during that window, if a planet really is making regular transits.
Why a dip in brightness isn’t a planet yet
Some context keeps this grounded. Astronomers have already cataloged more than 6,400 exoplanets. Some orbit a star; others, so-called “rogue planets,” drift through space with no gravitational anchor, true to the Greek root of “planet,” which means “to wander.” Billions more are believed to be out there. In eight years of scanning, TESS has identified more than 1,000 exoplanet candidates that astronomers later confirmed.
And the method has well-known traps. A binary star system, for example, can easily be mistaken for a planet passing in front of a star.
“The biggest challenge with confirming planets is that we have to rule out false-positive scenarios,” said Kevin Hardegree-Ullman, a research scientist at the NASA Exoplanet Science Institute (NExScI). “This is difficult because: (1) there are just so many candidates… (2) we have limited telescopes… and we are generally competing with all other astronomers for time on these… and (3) some targets are not amenable to follow-up measurements with current telescopes (e.g., they are too faint to get a good signal other than a transit-like event).”
Rabtsevich doesn’t oversell it either. He stresses that correlation doesn’t necessarily mean causation, and the pattern may turn out to come from something other than a planet.
Where AI fits, and where it doesn’t
Detecting candidates is the part AI handles well. Confirming them is a different job.
An AI model NASA deployed a little under a year ago, built to sift TESS data, has already identified around 7,000 exoplanets. Because tools like Claude Code and Codex are publicly available, they may also open the search to amateurs like Rabtsevich, far outside the scientific establishment.
“AI tools won’t necessarily be helpful in the process of confirming planets,” Hardegree-Ullman said, “but [they] will likely continue to be used to string together publicly available code to help search for planets and run them through the basic vetting processes… ”
For Rabtsevich, the speed was the whole story.
“Without these tools I couldn’t have processed a data set this size and carried it through to a result,” he said. “Even a couple of years ago my job meant building endless spreadsheets and grinding through data sets, and that ate up a huge amount of time. Here, a search I planned went through 126,000 stars, and then dozens of tests ran on a single one, in about two weeks.”
The flood he’s worried about
His post has already pulled other amateurs into the hunt. That’s not purely good news, and Rabtsevich knows it.
“Since my [X] post, I’ve seen a lot of people start searching TESS data for planets themselves,” he said. “I hope that enthusiasm comes with careful checks, so professional astronomers aren’t left sorting through poorly vetted signals.”
Given Hardegree-Ullman’s point about scarce telescope time, that warning carries weight. A wave of half-checked AI candidates would cost professionals the exact resource they’re shortest on.
Not everyone is reading it that cautiously. In a Claude subreddit, one Redditor responded to Rabtsevich’s findings with relief at an AI story that wasn’t about slop: “There is an awful lot of absolute shite made with the use of AI,” the user wrote. “But this… this is fucking incredible. Well done.”
Hold the applause until November. If TIC 4206066 dims on the schedule Rabtsevich posted, between October 31 and November 26, his process will have earned it. If it doesn’t, he’s promised to say so, and that promise is the most credible thing in this whole story.
