Meta is grading every employee on AI-driven impact this year

Meta will grade every employee on AI-driven impact in 2026 performance reviews. WinBuzzer reported on 4 February that the company is the first major tech firm to write AI into formal evaluation criteria, following an internal memo from Head of People Janelle Gale that Business Insider reported in November. Gale's line: "For 2025, we'll reward those who made exceptional AI-driven impact, either in their own work or by improving their team's performance." The policy covers all roles, engineers to marketers. Engineering managers will assess whether workers used AI to accelerate development cycles and improve code quality. Meta has been tracking usage through internal dashboards. This is not confined to one company: a Microsoft executive told managers last June that using AI is no longer optional, and Sundar Pichai told Google staff at a July all-hands that they need to use AI.

The scored unit is attribution, not adoption

Read the memo language precisely. Meta is not promising to reward the highest token count. It is rewarding "AI-driven impact" — an outcome with a causal claim attached to it. Those are different metrics, and the gap between them is where careers are decided this year.

The reason the impact framing wins is that the usage framing does not survive contact with the finance function. Gartner research cited in the same reporting found that only one in 50 AI investments delivers transformational value, and only one in five delivers any measurable return. Gartner also found that less than 1% of layoffs in the first half of 2025 resulted from AI productivity gains. Dashboards full of active seats have not been converting into defensible numbers.

One finding does convert. Gartner reports that business units which redesign work processes around AI are twice as likely to exceed revenue goals. Redesign, not assistance. That distinction is the whole brief. A worker who uses a model to draft faster produces a slightly cheaper version of the old process. A worker who removes three steps from a workflow and can show the cycle time before and after produces something a manager can defend in a calibration meeting.

The consensus reading of Meta's policy is surveillance — another dashboard watching another worker. That reading is comfortable and wrong. Surveillance data is thin evidence and everyone above the review knows it. The scarce input is a credible causal story about a changed process, and that story has to be supplied by the person who changed it.

The unscored cycle is where the standard gets set

Meta phased this deliberately. Individual AI usage metrics were kept out of the 2025 reviews, but employees were urged to highlight AI-powered wins in their self-reviews, with bonuses or raises available as special recognition. Internal surveys had already shown that parts of Meta's own engineering workforce were not using the available tools consistently.

Treating that as a grace period is the error. The submissions filed in the unscored cycle become the reference examples for what "exceptional AI-driven impact" means when scoring starts. Whoever wrote a strong entry defined the bar. Everyone else now gets measured against a standard set by colleagues who were paying attention a year earlier.

The same structure is arriving in ordinary companies, minus the memo. Gartner found that 81% of CIOs say AI skill gaps impede their ability to meet objectives, which is exactly the pressure that produces adoption mandates. And the appetite for machine-assisted judgment already exists inside the workforce: in an earlier Gartner survey, nearly 90% of employees said algorithms could give fairer feedback than their managers.

There is a real cost to compliance without thought. Roberto Rigobon, Professor of Applied Economics at MIT Sloan, has warned that outsourcing cognitive tasks erodes the underlying skills, drawing the parallel to forgetting calculus once you stop using it. Being graded on AI usage and quietly deskilling at the same time is a live possibility, not a hypothetical.

Your Next Move

Open an impact log this week and keep it to one line per week. Record the process you changed, the tool you used, the measurable before and after, and who else benefited. Rheanne Boren, Director of Global Talent Management at Micron Technology, noted that without tracking, workers reach the end of the year and forget the good work they did. Recency bias is not fixed by memory; it is fixed by receipts.

Redesign one process this month rather than assisting ten. Pick something you own end to end, cut steps out of it, and record cycle time on both sides of the change. That is the artefact that maps onto the only Gartner finding with revenue attached to it.

Keep one core task deliberately unassisted. Choose the judgment your role is actually paid for and do it without a model, weekly. The review measures what AI added. Your market value still rests on what you can do when it is switched off.