PM job is changing in AI firms
Product management in fast-growing AI companies is shifting from writing specs to building with engineers and AI tools side-by-side. Social posts highlight Anthropic as a high-learning place for PMs due to rapid ARR growth and talent density, and leaders note that PMs are now prototyping in models like Claude instead of producing long PRDs (x.com) (x.com). That shift collapses feedback loops and raises the bar for technical fluency and rapid decision-making inside PM teams (x.com).
The old product manager job in software was to write a long document, hand it to engineers, and wait a week to see what broke. In artificial intelligence companies, that week is collapsing into an afternoon because the product manager can now ask a model to build the first version on the spot. (developers.openai.com) That change is showing up most clearly at Anthropic, where the company’s own hiring pages say candidates should already know how to collaborate with Claude and where its Claude Code product manager role asks for both product management experience and at least 1 year as a professional engineer. The same role pays $285,000 to $305,000 a year and says the product manager should be able to debug workflows with engineers, not just write requirements for them. (anthropic.com 1) (anthropic.com 2) Anthropic is hiring against a business that has been growing at a speed that barely existed in software a few years ago. Sacra estimated Anthropic reached $30 billion in annualized revenue in March 2026, up from $9 billion at the end of 2025, which helps explain why people in product see it as a place where decisions get stress-tested fast. (sacra.com) The workflow itself is different now. OpenAI’s developer guide says coding agents can scaffold projects, generate files, run tests, and carry a feature from proposal to deployment, which means a product manager can check an idea with working software instead of arguing over a slide deck. (developers.openai.com) That removes one of the oldest bottlenecks in product work: the translation step between “what users need” and “what engineering should build.” When a model can turn plain English into a prototype, the product manager no longer needs a 20-page product requirements document to prove the idea is coherent. (developers.openai.com) (builder.io) You can see the market adapting around that behavior. Builder.io’s March 2026 guide for product managers pitches Claude Code as a way to turn plain-English requests into working prototypes, analyze comma-separated value files, and connect with tools like Jira, Slack, and Linear, all jobs that used to be split across several people and several meetings. (builder.io) The skill bar moves with the tools. Anthropic’s Claude Code product manager listing says the person should be “equally comfortable” discussing feedback with researchers, debugging with engineers, and presenting strategy to executives, which is a much narrower target than the older version of product management that rewarded polished documents and stakeholder management alone. (anthropic.com) The speed change is not subtle. OpenAI says development cycles that once took weeks are now being delivered in days, and when the build-test loop gets that short, the person deciding what to try next needs technical judgment in real time instead of a talent for writing perfect specifications in advance. (developers.openai.com) That is why fast-growing artificial intelligence firms are starting to look less like companies with product managers and engineers in separate lanes. They look more like small mixed teams where the product manager sits beside the engineer, opens Claude or another coding agent, and helps ship the first draft before the meeting is over. (anthropic.com) (developers.openai.com)
