Meta redefines AI evaluation: focus shifts from adoption to output

Meta engineering leaders Maher Saba and Santosh Janardhan issued a memo to employees, stating the company will no longer use AI adoption dashboards or token metrics to assess employee impact. The memo highlights that 93% of code changes at Meta are now aided by AI agents, with AI deeply integrated into daily engineering workflows. However, the memo stresses that AI adoption is not the ultimate goal. Industry trends over the past two years have seen companies track metrics like code generation ratios, Copilot adoption rates, and token consumption to promote AI tool use. Some linked these metrics to performance evaluations, leading to 'performative AI use' focused on quantity. By setting boundaries, Meta separates tool usage intensity from actual output, returning employee impact evaluation to delivery results. Meta's 93% AI agent penetration rate signals that AI programming tools have reached an infrastructure level in large engineering systems. Evaluating 'whether it is used' now holds little value. Shifting focus to output quality aligns with the industry trend of moving from promotion to practical results after widespread AI adoption.