Ethan Carter
Imagine working in a peculiar role where your day revolves around firing guns at targets. Initially, you were informed that your worth to the organization would depend on the quantity of bullets you discharged, prompting you to pull the trigger excessively. Then, just as a bullet shortage became apparent, you received a memo indicating that there was no longer a need to fire so many bullets. But coincidentally, around the same time, you were handed a minigun reminiscent of the one from Terminator 2: Judgement Day.
According to Wired, a scenario reminiscent of this analogy is unfolding with AI at Meta.
A March column by Kevin Roose in the New York Times introduced many to the concept of “tokenmaxxing,” where AI-centric workplaces, including Meta and OpenAI, sought to maximize AI integration among employees. Roose hinted at the then-popular AI platform OpenClaw, which may have fueled this trend. He noted that Meta was “rewarding workers who heavily utilized AI tools while admonishing those who did not.” Leaderboards were mentioned, turning the overuse of AI into a competitive endeavor.
A few months later, in June, the Financial Times reported that Google was limiting Meta’s AI tokens, compelling the company to reassess its approach to tokenmaxxing. It appears that tokenmaxxing is now officially being phased out. A report from The Information revealed that Meta engineers had been informed that “AI adoption dashboards” and “token counts” would no longer factor into performance evaluations.
Wired, citing anonymous insiders from Meta, indicates that newly released performance review guidelines no longer include “AI usage” as a measurement criterion. The new guidance states that desired outcomes “can be supported by AI or other means.” Sources within Meta have expressed relief at the removal of pressure to constantly engage with AI.
Simultaneously, Meta has been working on developing a competitive AI tool of its own.
In 2024, Mark Zuckerberg proclaimed, “This will be the year when a highly intelligent and personalized AI assistant reaches over 1 billion people, and I expect Meta AI to be that leading assistant.” However, it remains unclear whether Meta’s ChatGPT-style application ever achieved this goal. Personally, I used the Meta AI web app just once for a story. Amanda Siberling from TechCrunch also tried Meta AI for a story but had a less favorable experience; Instagram notified her friends of her usage, leading to some teasing.
It appears that Meta is now aiming for better outcomes with a productivity-focused tool rather than one designed to compete directly with ChatGPT. In May, The Information reported that Meta was developing a project named Hatch, intended to be Meta’s own version of OpenClaw—an approachable AI platform similar to the kind that originally fueled the tokenmaxxing phenomenon.
According to Wired, alongside the apparent discontinuation of tokenmaxxing incentives, Meta “has been promoting—but not mandating—employees to utilize its most advanced AI experiment.” This experiment is Hatch, which has yet to be released but has been accessible to employees “for the past several weeks,” as Wired reports. This tool reportedly consumes more AI tokens than standard chatbots and AI coding tools. While some employees enjoy using Hatch, others recall an incident from April when it was revealed that Meta planned to leverage employee keystrokes to train AI, causing apprehension about the new tool. (Meta subsequently paused that initiative).
Heavy reliance on AI agents remains ingrained in Meta’s workplace culture. A memo issued earlier this week by AI chief Alexandr Wang explained that the company’s transition from Google Chat to Slack was designed to enhance the use of AI agents. Wang stated that Slack is “the most robust platform available today for agents.” He emphasized that while “not all of us are developing agents today,” the entire company would benefit from a strong agent ecosystem.
If Hatch succeeds internally, Wired expresses concern that layoffs may soon follow. A recent Reuters report indicated that Meta planned to reduce its workforce by up to 60% in specific areas. the plan did not progress due to glitches in the AI software intended to replace the laid-off employees.
Gizmodo reached out to Meta for comments regarding the Reuters report and the internal testing of Hatch. Unfortunately, no response was received.
Daniel Mercer
