Mark Zuckerberg wants Meta’s engineering teams to restructure entire codebases—not to fix bugs or ship features, but so AI agents can read, navigate, and modify the code themselves. According to a report by The Information, the directive was described internally as a “bold ask,” and it signals just how far Zuckerberg is willing to push his AI-native vision for the company.The mandate fits into a broader pattern at Meta, where AI adoption is no longer optional. The company has set explicit targets for how much engineers should rely on AI tools, reorganised teams around small AI-focused “pods,” and started tracking employee AI usage on internal dashboards. CTO Andrew Bosworth recently took charge of Meta’s internal “AI for Work” initiative, according to Business Insider.
Meta is tracking AI usage across the company—down to the token
The push doesn’t stop at code. According to The Information, Meta now maintains an internal dashboard that logs token consumption—a measure of how much data employees process through AI tools—across different roles. Over a recent 30-day period, total token usage reportedly crossed 60 trillion tokens.That number carries real cost implications. Using Anthropic’s public pricing as a rough benchmark, that volume of usage could translate to around $900 million a month. Meta’s actual costs likely differ based on internal infrastructure and custom pricing deals, but the scale tells its own story.Employees have also turned token usage into a competition. Badges like “Token Legend” and “Session Immortal” are awarded on a leaderboard internally called “Claudeonomics”—a nod to Anthropic’s Claude models—that ranks the top 250 users out of more than 85,000 employees.
Inside Meta’s AI-native playbook: pods, titles, and hackathons
The cultural shift runs deeper than dashboards. Business Insider reported that a 1,000-person team within Reality Labs has been restructured into AI-native “pods,” with employees given titles like “AI Builder” and “AI Pod Lead.” The reorganisation was designed to flatten Meta’s hierarchy and encourage cross-functional work—engineers doing design, for instance, if that’s what the project needs.Meta has also been running intensive “AI Transformation Weeks,” where employees across roles attend hackathons, watch demos, and build with tools like Anthropic’s Claude Code. An internal document reviewed by Business Insider showed that Meta’s creation org set a first-half 2026 goal for 65% of engineers to write more than 75% of their committed code using AI assistance.
More tokens, but is the work actually better?
Not everyone inside Meta is convinced the approach is working as intended. Some employees have reportedly extended AI sessions or run unnecessary tasks just to boost their visible token usage—a behaviour that prioritises engagement metrics over actual output.Meta has not formally tied token consumption to performance reviews. But the pressure to demonstrate AI fluency is hard to miss. As one employee put it to The Information, there’s a meaningful gap between using AI well and simply using a lot of it.
