Peter Bailis joins Anthropic

Peter Bailis left Workday’s chief technology officer job after less than a year and turned up at Anthropic as a “member of technical staff,” a title that sounds smaller on paper but sits much closer to the model-building work. Anthropic confirmed he joined to work on reinforcement learning engineering, the part of artificial intelligence training that teaches a model by rewarding better behavior over many runs. (thenextweb.com) The timing is the point. Workday hired Bailis in May 2025 to help push the company “all in” on artificial intelligence, and by March 2026 he had left for a frontier model lab that is now moving toward the same enterprise workflows Workday sells into. (thenextweb.com) Workday is not a consumer app company. It sells payroll, finance, and human resources software to large employers, which means its systems hold the neat rows and columns of company life: job titles, compensation bands, approvals, budgets, and org charts. (theinformation.com) Anthropic is coming from the other direction. Its enterprise pitch for Claude is that companies can securely connect the model to internal knowledge, so the model is not just chatting in a vacuum but answering with company context attached. (claude.com) That is where Bailis’s background fits. Before Workday, he founded Sisu Data and earlier taught computer science at Stanford, building a reputation around data systems, analytics, and the machinery that turns messy business records into something software can reason over. (bailis.org) He was also publishing on compound artificial intelligence systems as recently as 2025, including work on how to choose among multiple models inside one larger system. That is the kind of problem you hit when one model writes, another retrieves data, and a third checks the answer before anything reaches a paying customer. (openreview.net) So this hire is not just Anthropic adding another senior name. It is Anthropic adding someone who has spent years on the unglamorous layer between a clever model and a real company database, where permissions, schemas, and bad records usually break the demo. (bailis.org) The “member of technical staff” label also tells you something about where power is moving. At frontier artificial intelligence labs, senior people are increasingly leaving executive titles for hands-on technical roles, because the leverage now sits with the teams tuning models, building agents, and wiring those agents into business systems. (finance.yahoo.com) Anthropic’s move into human resources software makes the overlap even sharper. If a model company wants to help with hiring, performance reviews, or internal support, it needs more than fluent text generation; it needs reliable access to the structured fields that companies already store in systems like Workday. (thenextweb.com) That is why this looks less like one executive changing jobs and more like a border crossing. The people who know how enterprise data is stored, cleaned, and governed are becoming as valuable to model labs as the people who know how to train the models themselves. (erp.today)