When Silicon Valley races in the future, in which AI agents perform most of the software programming, a recent problem arises: finding errors generated by AI before entering them into production. Even OpenAI deals with such problems, described by a former worker.
Newly financed startup Playero He created a solution: Use AI agents trained to find and repair problems before entering the code for production, informs TechCrunch, CEO of the startup and founding father of Sole, Animesh Koraratana.
Koraratana created Playero when he was in Stanford Dawn Lab for Machine Learning under his adviser and founding father of the laboratory, Matei Zaharia. Zaharia is after all the famous programmer and co -founder of Databicks; He created his fundamental technology while working on his own doctorate.
On Wednesday, Playero announced that he collected a series A in the amount of $ 15 million and led by Ashu Garg from the Foundation, an early supporter of information. Seeds with a value of $ 5 million run by Green Bay Ventures and a few noteworthy angels, including Zaharia, general director of Dropbox Drew Houston, general director of Figma Dylan Field and CEO Vercel Guillermo Rauch.
During his stay in Stanford Dawn Koraratana, currently 26 -year -old, he worked on the compression technology of the AI model and “was really early exposed to language models,” he says. He met programmers who created some of the first AI coding tools.
He was struck then that “there is this world in which the computers will write the code. It will no longer be people,” said Koraratana Techcrunch. “What will the world look like at the moment?”
He knew that before the term “AI Slop” he was even invented that these agents intend to produce a code that broke things like their supervisors.
TechCrunch event
San Francisco
|.
October 27-29 2025
This problem would even be tightened by so many agents who develop much more code than ever before. People won’t all the time be practically checking all the code written by AI for errors or hallucinations. And the problem becomes much more intense for large, complex code bases on which enterprises rely.
Playerzo trains models “that really deeply understands the bases of the code and we understand the way they are built, the way they are archived,” says Koraratana.
His technology studies the history of errors, problems and solutions of the company. When something bursts, his product can “find out why and fix and then learn from these mistakes to prevent them from repeating them again,” says Koraratana. It compares his product to the immune system for large code databases.
Landing Zaharia, his adviser, as an angel was the first step to collecting funds, but the moment that actually confirmed his idea was when he showed a demo to one other famous developer: Rauch. Rauch is the founder Triple Unicorn Developer Tool Company Vercel and creator of the popular Open Source JavaScript Framework (*15*)Next.js.
Rauch watched Koraratana’s demo with interest, but skepticism, asking how many of them were “real”. Koraratana replied that it was a code “in production. This is a real example. And he was quiet,” says Koraratana. Then his future investor replied: “If you can solve it in the way you imagine, it’s really a big deal.”
Of course, the playero is not alone, trying to solve the problem of the error generated by AI. Last week, the Anysphere cursor Bugbot was launched Detect coding errors as one example.
Despite this, the playero is already gaining grip on the emphasis on large code databases. Although it was developed for a world where agents are coders, it is currently used by several large enterprises that use the coding of the second pilots. For example, the company’s invoicing company is one of the clients of the startup awning. Zuora uses technology in its engineering teams, including to observe its most respected code, its billing systems.