Meta is quietly walking back the internal pressure campaign that pushed employees to rack up AI tool usage, even as it doubles down on getting staff to test Hatch, its most ambitious AI agent yet. The shift, first reported by Wired, signals a company recalibrating how it measures AI adoption internally while still racing to prove its agent technology can compete with rivals.
Meta is backing off a push that had employees feeling pressure to rack up usage numbers on the company's internal AI tools, even as it leans harder into getting workers to try out Hatch, described as the company's most advanced AI agent project yet, according to Wired. It's a subtle but telling shift for a company that has spent the last two years trying to prove, both to Wall Street and to its own workforce, that generative AI belongs in every corner of the business.
For months, Meta had reportedly encouraged, and in some cases nudged, employees toward what insiders call 'tokenmaxxing,' essentially running up usage of internal AI coding and productivity tools as a way to demonstrate adoption. That kind of top-down usage pressure isn't unique to Meta. Tech companies across the industry have leaned on internal dogfooding campaigns to build confidence in their own AI products before pushing them to customers. But forcing adoption through soft mandates can backfire, producing inflated usage metrics that don't reflect real productivity gains, a concern that appears to be driving Meta's course correction now.
Instead, the company's energy seems to be moving toward Hatch, the AI agent that Meta is treating as its next big internal proving ground. While the specifics of what Hatch can do haven't been fully detailed publicly, the framing as the company's 'most advanced AI project yet' suggests it's meant to go well beyond simple chatbot functionality into more autonomous, task-completing territory, the kind of AI agent technology that Google, OpenAI, and Microsoft have all been racing to ship over the past year.
This comes at a pivotal moment for Meta's AI ambitions. The company formed its Superintelligence Labs division in 2025, pulling in high-profile researchers and pouring resources into frontier model development after facing criticism that its Llama family of models had fallen behind competitors on key benchmarks. Mark Zuckerberg has repeatedly framed agentic AI, tools that can take actions on a user's behalf rather than just answer questions, as the next major battleground, telling investors on past earnings calls that Meta intends to be a leader in the space rather than a fast follower.
The decision to ease pressure on general AI tool usage while concentrating internal enthusiasm on Hatch specifically suggests Meta wants cleaner signal on what's actually working. Blanket usage mandates tend to generate noisy data. If everyone's told to use an AI tool regardless of whether it helps their workflow, the resulting metrics say more about compliance than genuine utility. By contrast, funneling curiosity toward a single flagship project like Hatch gives Meta's product teams a more focused testbed, and arguably a more honest one, for gauging whether the agent actually delivers value before any wider rollout.
It also reflects a broader industry reckoning. Companies from Salesforce to Google have faced internal grumbling over AI mandates that employees felt were more about optics than genuine efficiency gains. Workers pushed to hit usage targets have, in some cases, found workarounds that technically satisfy metrics without meaningfully changing how they work, the exact kind of gamed adoption data that makes AI ROI claims harder to trust externally.
For Meta, getting Hatch right carries real stakes. The company has poured billions into AI infrastructure and talent, and investors are watching closely for signs that the spending translates into products, whether consumer-facing features in the Meta AI app or enterprise tools, that actually move the needle. An internal agent that genuinely earns organic adoption from employees would be a far stronger signal to the market than usage stats built on mandates.
What happens next will say a lot about Meta's confidence in Hatch. If the agent graduates from internal testing to a public or developer-facing release, expect the company to lean heavily on its origin story: not a tool employees were forced to use, but one they reportedly wanted to.
The quiet retreat from forced AI usage metrics, paired with a concentrated push behind Hatch, hints at a more mature phase of Meta's internal AI strategy, one focused on proving genuine value rather than just juicing adoption numbers. For employees, it likely means less pressure to perform AI enthusiasm and more room to judge Hatch on its actual merits. For the industry, it's a reminder that mandated usage isn't the same as real product-market fit, even inside the company that built the tool.
