Meta tried to shrink engineering teams around AI (leaddev.com)

Meta's recent initiative to restructure its engineering teams around AI has yielded mixed results. The company experimented with "pods" of three to five workers, leading to a remarkable 220% increase in code changes year-on-year. However, the benefits did not translate effectively into usable features, which only increased by 36%, while incidents of technical and security problems surged by 40%. This revealed a critical oversight in Meta's approach: while AI can enhance output speeds, it does not inherently improve the quality or utility of the work produced. The situation reflects a broader trend in the engineering community, where metrics like code throughput may mask underlying issues in developer experience. Research from developer intelligence firm DX indicates that while PR throughput has risen across multiple organizations, overall developer satisfaction has declined. Experts suggest the ideal team size lies between five to eight members, as smaller teams may struggle with the influx of AI-generated changes, exacerbating existing challenges. Leaders are urged to rethink team structures and adopt a gradual approach to integrating AI, emphasizing the importance of meaningful output over mere code volume. Meta's experience underscores the complexities of melding AI technology with human-centric engineering practices effectively.