AI Is Now Writing Your Performance Review — How Meta's Formal Impact Scoring Is Setting the Corporate Template
By Tim Kreling, Co-Founder, OVI
For decades, the annual performance review has followed the same arc: a manager recalls the past twelve months, a rating is assigned, and the employee learns their fate. That model is crumbling — not because companies decided it was broken, but because AI made something better possible. Meta's 2026 policy change is the starkest example of where corporate performance management is heading.
In November 2025, Janelle Gale, Meta's Head of People, sent an internal memo to all employees: starting in 2026, "AI-driven impact" would become a core expectation across every role, from engineers to marketers, and that impact would be measured, tracked, and incorporated into formal performance ratings that affect promotions and bonuses. It was not a suggestion to use AI more. It was a structural redesign of how the company evaluates its people.
What Meta Actually Built
The policy is not simply a checkbox on a review form. Meta built an interconnected infrastructure to make AI usage measurable.
Checkpoint is the engine. It is an AI-powered performance tracker that aggregates more than 200 data points for software engineers, including the ratio of AI-generated code to manually written code, usage frequency of Meta's internal coding assistant, and contribution velocity. For non-engineering roles, Checkpoint tracks engagement with productivity bots, use of Metamate (Meta's internal LLM assistant), and adoption of AI-powered workflow tools across marketing, operations, and product.
Level Up is the on-ramp. Before formal evaluation begins, Meta launched a gamification layer that rewards employees with digital badges as they hit AI milestones. Completing AI-assisted projects, reaching usage thresholds with Metamate, and experimenting with new tools all earn points. The goal was to build familiarity before the stakes were formal — and to create a cultural signal that AI adoption was expected, not optional.
The AI Performance Assistant is the review support layer. Integrating both Metamate and Google's Gemini, it gives employees access to AI tools specifically for preparing their self-reviews, surfacing data from their own Checkpoint history to support their self-evaluations. The irony is intentional: employees now use AI to argue their case for how well they used AI.
The result is a performance management system where AI usage is quantified with more granularity than most companies apply to any other work behavior.
Why This Matters Beyond Meta
Meta's approach is not an isolated experiment. It is the leading edge of a wave that SHRM data has been tracking. The organization's 2026 State of AI in HR report found that 42.3% of organizations are now using AI in their talent strategies — more than double the 17.9% in 2025. Of HR leaders surveyed, 33.7% said they plan to invest in AI-driven performance management tools in the next 12 months, making it the top-funded category in the survey.
The direction is clear. What is less clear is how most companies are handling the transition. A May 2026 Forbes analysis found that most managers have not updated their evaluation frameworks to match AI-era expectations. They are still assessing employees on outputs defined before AI tools existed, using rubrics that do not account for whether results were produced independently or through AI augmentation.
Meta chose to resolve that ambiguity with explicit policy. Others are arriving at similar conclusions through different paths.
How Unilever Approaches the Same Problem
Unilever's strategy shares the same underlying goal — making performance management more responsive and data-informed — but takes a less quantified approach. The company introduced an AI-powered continuous feedback system that analyzes employee performance data in real time, identifies patterns across multiple check-ins, and generates tailored development suggestions automatically.
Where Meta tracks specific tool usage, Unilever tracks outcomes and feedback signals over time. Where Meta uses gamification to drive adoption, Unilever uses real-time coaching prompts to keep development conversations active between formal review cycles. Both approaches acknowledge the same problem: annual point-in-time evaluations are too infrequent and too subjective to drive performance in an AI-augmented workforce.
The difference in approach reflects different organizational philosophies, but the destination is the same — a performance system that uses AI both to evaluate work and to improve the quality of that evaluation.
The Risks HR Leaders Cannot Ignore
No honest assessment of this shift can skip the risk register.
Gaming the system is the first-order problem. When AI usage becomes a rated metric, employees will optimize for the metric. Developers may run AI-assisted code generation on routine tasks they would have handled faster manually, simply to maintain their Checkpoint ratio. Marketers may route work through Metamate that does not benefit from AI processing. The signal becomes noise the moment it becomes a target.
Meta's 200-point measurement system is an attempt to make gaming harder, but it cannot fully solve the problem. Any system that measures behavior changes the behavior it measures.
Fairness and bias are the second-order problems. Not all roles benefit equally from AI tools. Employees in client-facing or highly collaborative functions may have fewer natural opportunities to demonstrate AI usage than employees in production or analytical roles. Roles that require human judgment — sensitive HR decisions, legal assessments, executive communications — may be disadvantaged in a system that rewards AI augmentation.
Surveillance concerns are the third. Checkpoint aggregates more than 200 data points. Employees reasonably ask what else is being tracked, who has access to the data, and how long it is retained. In jurisdictions with robust employee privacy protections — the EU and several US states — companies implementing similar systems will need to address legal questions about monitoring and consent that Meta has not fully addressed publicly.
What HR Leaders Should Consider Now
The Meta model will not be appropriate for every organization, but the underlying pressure is universal. AI capabilities are changing what "good work" looks like, and performance management systems that predate this shift are increasingly measuring the wrong things.
Start with role clarity before you build metrics. The first step is not purchasing a Checkpoint equivalent — it is mapping which roles are plausibly improved by AI and which are not. A performance system that penalizes employees in human-intensive roles for low AI usage will damage trust and produce worse outcomes than the system it replaced.
Distinguish between AI augmentation and AI dependency. Some organizations are already discovering that heavy AI reliance can degrade underlying skill development, particularly for early-career employees. A performance framework that rewards AI usage without tracking whether outcomes actually improve will optimize for tool engagement, not for competence.
Build the culture before building the system. Meta used Level Up to normalize AI adoption before making it a formal metric. The sequence matters. Organizations that introduce AI usage as a rated dimension without first building genuine capability and comfort will encounter resistance that the measurement system itself cannot resolve.
Consult legal before you deploy tracking. Particularly for companies with EU workforces, the data collection requirements of a system like Checkpoint may require Data Protection Impact Assessments, works council consultations, and disclosure obligations that take months to process. Design the legal framework before committing to the product.
The performance review is not disappearing — it is being rebuilt around different inputs. The organizations that navigate this transition well will be the ones that decide deliberately what they want to measure and why, rather than adopting measurement infrastructure because it is available.
Has Meta publicly confirmed the specifics of its Checkpoint system?
Meta has not released a comprehensive technical specification of Checkpoint. Details about its 200+ data points and AI code ratio tracking have been reported by HR Grapevine, eWeek, and technology journalists citing internal sources and company announcements. The broad policy — that "AI-driven impact" is now a formal review metric — has been publicly confirmed by Meta's Head of People.
Which other major companies are formally tying AI usage to performance ratings?
JPMorgan implemented a mandate requiring employees to demonstrate AI tool adoption as part of how their work is evaluated. Unilever has deployed a continuous AI feedback system that informs development conversations. Several major consultancies and financial services firms have introduced informal expectations around AI tool usage. Meta's approach is currently the most formalized and publicly documented.
How should HR teams handle employees who resist AI tools?
Resistance often reflects legitimate concerns — skills erosion, surveillance, or lack of genuine utility in their specific role. Before treating resistance as a performance issue, HR teams should audit whether the tools being required actually improve outcomes in the role in question. Mandating AI usage in roles where it adds no value creates compliance behavior that looks like adoption but delivers none of the productivity gains.
What is the legal risk of tracking employee AI usage?
In the EU, systematic monitoring of employee tool usage may trigger GDPR obligations including Data Protection Impact Assessments and transparency requirements. In the US, state-level employee monitoring laws vary significantly — California, Connecticut, and New York have the most stringent requirements. Companies should complete a legal review before deploying usage-tracking infrastructure similar to Checkpoint.
Can smaller companies implement a version of this approach without Meta's infrastructure?
Yes, at a simpler level. Many existing performance management platforms — including Betterworks, Culture Amp, and Lattice — now include AI usage signals as an optional data layer. The principle is accessible without custom infrastructure. The more important question is whether the organization has defined clearly what AI augmentation should look like in each role before introducing it as an evaluation criterion.
When does measuring AI usage in performance reviews cross into micromanagement?
The line is outcome versus behavior. Measuring whether employees are achieving better results — faster, higher quality, more consistent — with AI tools available to them is outcome-based and defensible. Measuring whether they clicked a specific button a required number of times per week is behavior-based and will be perceived as surveillance. Systems like Checkpoint sit closer to the behavior end of that spectrum, which is a distinction worth naming clearly before deploying any similar infrastructure.