LeadDev reports that Meta tested engineering pods of three to five people, each using coding agents. Its September 7 article says the teams produced 220% more code changes year over year but delivered only 36% more user-facing features.
LeadDev also reports 40% more major technical and security incidents and 70% more time spent on firefighting. The story does not publish the underlying Meta dataset, so those figures should be treated as reported company metrics rather than independently audited results.
The article places Meta's experiment alongside research from developer-intelligence company DX. Across more than 500 organizations, DX found a 37% rise in median weekly pull-request throughput over four quarters while its developer-experience index fell from 67 to 65. That survey covers different teams and measures; it is not a direct validation of Meta's results.
The account points to pressure on code review, testing and operations when AI tools make generating changes faster. Engineers interviewed by LeadDev argue that teams need strong evaluation and deployment systems before reducing headcount. Their preferred team size of five to eight people is expert advice, rather than a finding established by Meta's experiment.