Meta's internal ChatGPT-style bot is helping employees with performance reviews (www.businessinsider.com)

Meta employees are using an internal ChatGPT-style assistant called Metamate to help write and summarize year-end performance reviews. Joseph Spisak, a product director in Meta Superintelligence Labs, said Metamate can search employees’ internal documents, extract accomplishments, and assemble summaries and feedback — effectively pulling together a portfolio of work when someone asks it to “summarize my performance.” Meta trains Metamate on internal data and deploys similar AI helpers across the company (from coding assistants like Devmate to dashboards and internal apps), and some staff say they use it to generate peer feedback by feeding templates and examples into the tool. The rollout highlights both the productivity upside and the limits of embedding LLMs into core HR workflows. Technically, Metamate’s strength is cross-document retrieval and synthesis over private corpora, but employees report uneven outputs when projects lack detailed context, raising accuracy and bias concerns. Spisak’s tongue-in-cheek “reward hack it” comment underscores governance risks: automated summaries could be gamed, misrepresent contributions, or reinforce existing evaluation biases if unchecked. For AI/ML teams, this is a real-world case study in model grounding, retrieval-augmented generation, and the need for human review, robust context windows, and policy controls when LLMs participate in personnel decisions.