DevOps / SRE / Platform · 05.08.2026, 18:24 UTC
The 800 mistakes that could reshape Meta’s AI coding strategy
| Schweregrad | info |
|---|---|
| Kategorie | DevOps / SRE / Platform |
| Quelle | The New Stack ↗ |
| Veröffentlicht | 05.08.2026 UTC |
Sicherheitsmeldung mit Schweregrad noch nicht bewertet. Technische Details im Tab „Originaltext“; empfohlene Schritte in der Checkliste.
Meta is asking thousands of its software engineers to help train its internal AI coding tools by simply fixing code, something the team does all day.
Last month, Maher Saba, vice president of Meta’s Applied AI Engineering organization, asked engineers across the company to submit at least one code “diff” each week using MetaCode, the company’s internal AI coding agent, according to an internal memo obtained by The Information.
Saba said the corrections that come from engineers fixing a bad batch from MetaCode have already helped improve Muse Spark 1.1 and will be used to post-train an upcoming model known internally as Watermelon. According to the memo, 7,000 weekly active users have already submitted more than 800 fixes so far. Meta has even added colored badges to employees’ internal profiles based on the number of changes they contribute to encourage them to find more.
MetaCode gives Meta access to that entire process, which gives the company the ability to see the original task and MetaCode’s response, followed by the engineer’s correction and whatever tests or reviews were needed to approve it.
Engineers fix what AI breaks
Public repositories contain enormous amounts of working software, but they rarely capture things such as the model’s first attempt, where it went wrong, and what an experienced engineer changed before the code was ready to merge. But MetaCode gives Meta access to that entire process, allowing the company to see the original task and MetaCode’s response, followed by the engineer’s correction and whatever tests or reviews were needed to approve …