I spent two weekends doing something mildly unhinged. I pulled 200 product manager job descriptions from companies actively hiring on LinkedIn, Otta, and Wellfound, across big tech, growth-stage SaaS, seed startups, AI-native companies, and traditional enterprises. I scored each against a simple test: could this JD have been written in 2022, before the agent stack existed? 196 out of 200 could have been. Four were written for the role I actually run today. The other 98 percent are hiring for a job that is being automated as the JD is being posted.
The four tells of a dead JD
Every one of the 196 bad JDs had at least two of these, and most had all four. First, translation-layer language: "serve as the bridge between business and engineering." That describes exactly the work agents now do. Second, PRD-centric responsibilities: "write detailed PRDs." The PRD was a workaround for expensive engineering and expensive misunderstanding, and both premises are weaker now. Third, stakeholder management as the centerpiece: "manage cross-functional stakeholders." Coordination is exactly what a well-run agent pipeline eliminates. Fourth, the conspicuous absence: no mention of building, prototyping, evals, cost and latency, or any tool the role would touch daily. The absence is louder than the presence.
The four tells of a live JD
The four good ones shared four things. A specific prototype expectation in the first 90 days, not "contribute to product strategy" but "ship a working prototype of one customer-facing workflow, deployed to at least 10 users, in your first quarter." Named tools: Claude Code, Cursor, a chosen eval harness, a deployment stack, required, not "familiarity a plus." Owned metrics, not influenced ones, so the PM owns a retention or usage number rather than "helps drive" it. And an explicit disclaimer that PRDs are not the output, naming instead a prototype, an eval rubric, a deployed experience, a measured outcome.
The rewrite in three moves
If rewriting every JD from scratch is too much, here is the minimum-viable change, about 30 minutes each. Add a 90-day ship expectation to the top of the responsibilities, naming the artifact, the deployment target, and the measurable outcome. Name your stack under requirements as required, not preferred. Remove or rewrite any line describing translation, stakeholder management as primary function, or PRD-centric responsibility, and replace each with a verb that starts with ship, build, measure, or own. My rewritten "Growth Product Builder" example asks the hire to move the 14-day activation rate by 3 percentage points in the first quarter or come back with a clear reason why the bet did not work, to prototype two to four experiments a month without waiting on engineering, and to have shipped code in the last 30 days.
Pull three of your current open JDs this week and run each through that three-move edit. Ship the rewrites by Friday. You will know you got it right when the rewrite reads half as long and twice as clear.