Meta AI Adoption Revealed: Behind the curtain at Meta, 3 ways AI is already changing work

Meta is doing something useful for the rest of us. It is getting further down the road to an AI-first workplace while most employers still work out where the road goes.

And that means we can watch.

Meta AI adoption has gone from encouragement to expectation in a year. The company has reorganised thousands of jobs around AI and pushed employees to become AI-native. It has started testing an autonomous agent called Hatch. And its AI chief, Alexandr Wang, has just moved everyone’s internal conversations from Google Chat to Slack.

That last decision sounds mundane until you read the reason.

In a memo reported by Business Insider, Wang told employees that Slack offered a strong conversational interface for agents, mature developer tooling and integrations. “While not all of us are building agents today,” he wrote, “everyone at the company will benefit from a robust and useful agent ecosystem.”

Meta is choosing where its people and its agents will both work. Everything else follows from that.

Any HR team will recognise the position. The technology decision lands first, and the people questions follow it around for the next year.

Three things inside Meta are worth watching, because each one shows a different edge of the AI-first workplace.

1. Meta AI adoption is changing what workplace tech gets bought

For years, collaboration platforms have been assessed around employees. Usability, search, integrations, security, accessibility, and whether anybody can find the message they were sent last Tuesday.

Meta has added another user to that requirements list.

Slack has been positioning itself as an “agent-first workspace”, where employees talk to agents in the channels where work already happens. Slack says custom agents built on it have grown 300% since January, which is the vendor’s own figure. Gartner offers a firmer number. It expects 40% of enterprise apps to carry task-specific agents by the end of 2026, up from under 5% a year earlier.

Meta is buying into that idea early.

If agents become commonplace inside firms, workplace technology may get selected on two tests at once. What employees need from it, and what the organisation’s emerging population of digital workers can do there. For HR, EX and internal communications teams, Meta AI adoption makes the tech stack a far more interesting conversation.

2. Meta’s employees are the test population, not just the recipients

Meta employees are also testing Hatch, the company’s autonomous agent.

This is a step beyond asking a chatbot to summarise a document. Hatch runs tasks across applications and the web. Reported capabilities include deep research, form filling, browser control and bookings, from restaurants to dog sitting. Wired reports employees have had access for several weeks, and that Meta is encouraging rather than requiring them to use it.

So Meta’s staff are an early test population for something most employers are about to meet. AI that takes action rather than producing content. Meta AI adoption now means employees meet the agent well before the policy arrives.

That changes the shape of the employee experience questions. What is an agent allowed to do on somebody’s behalf? Permission, visibility and accountability all need answers before an agent books the wrong thing in somebody’s name. And once agents post alongside colleagues in a channel, employees need a way to tell human activity from automated activity at a glance.

Those questions are moving from hypothetical to operational. We have argued that every AI rollout needs an employee correction window. Agents make that harder to postpone.

3. Measuring AI use is not the same as measuring AI impact

The most useful lesson from Meta AI adoption is one the company has already started undoing.

As it pushed employees to become AI-native, internal attention landed on usage. Chief people officer Janelle Gale told staff that AI-driven impact would be a core expectation in 2026. An internal leaderboard called Claudeonomics then ranked the top 250 token users across a workforce of more than 85,000. Winners got titles like Token Legend and Cache Wizard.

Andrew Bosworth, Meta’s CTO, described the incentive plainly. “It’s like, this is easy money. Keep doing it. No limit.”

The leaderboard came down two days after news of it broke.

Meta has since told engineers that AI adoption dashboards and token counts will not be used to judge their impact. The new guidance says the outcomes it wants can be supported by AI or by other means. Managers have been pointed at quality, velocity, problem complexity and scope instead. That detail sits behind The Information’s paywall, so treat it as reported rather than confirmed.

The distinction sounds obvious. Plenty of employers will make the same mistake anyway. The spend crackdown coming for AI budgets will only make the measurement question louder.

Tell employees that AI adoption counts, and they will look for a signal of what good adoption means. Prompts sent, licences activated, weekly active users, hours supposedly saved. All wonderfully measurable. None of them a measure of better work, and the rework agents create rarely shows up in the same dashboard.

Meta has moved on to the harder question of what actually changed because someone used AI.

Nobody at Meta has found the answer yet

Meta is choosing infrastructure around agents and putting increasingly capable agents into employees’ hands. It is also unwinding its own AI incentives when they produce the wrong behaviour. Most employers will meet versions of all three on their way to an AI-first workplace. Meta simply got there first, and we can benefit from watching.