# How to Hire an AI Product Manager (What Startups Get Wrong) Hiring an AI Product Manager is different from hiring a traditional PM — and most startups do not realise this until they have already made an expensive mistake. After shipping AI products and watching teams navigate this hiring process, I have seen the same three mistakes play out repeatedly. Here is how to avoid them. ## Mistake 1: Prioritising AI Knowledge Over Product Craft The most common mistake is hiring for AI expertise at the expense of product fundamentals. The logic seems reasonable: you are building an AI product, so you need someone who deeply understands AI. But an AI expert who cannot run a product process is a researcher or an engineer, not a product manager. Product craft — discovery, prioritisation, stakeholder communication, delivery cadence — is learned over years. AI domain knowledge is learned in months. It is much faster to teach a strong PM about LLMs than to teach an AI researcher about product management. What to look for: A track record of shipping products — real outcomes, not just processes. Revenue influenced. Products launched. Teams led through ambiguity. Then assess AI literacy on top of that foundation. ## Mistake 2: Using Generic PM Interview Processes Standard PM interview loops test for things like market sizing, PRD writing, and prioritisation frameworks. These skills matter, but they do not differentiate an AI PM candidate from a generalist PM candidate. To assess AI PM readiness, add AI-specific scenarios to your interview process: **Scenario: Evaluation design.** Ask the candidate to define what "good output" looks like for an LLM feature in your product — and how they would measure whether the feature is meeting that standard. Strong candidates define clear, measurable evaluation criteria. Weak candidates describe the feature but not the measurement. **Scenario: Stakeholder communication.** Ask the candidate to explain a model accuracy tradeoff to a non-technical stakeholder who wants the feature shipped by end of quarter. Strong candidates find a clear framing without dumbing it down. Weak candidates either over-explain the technical details or avoid the tradeoff entirely. **Scenario: Prompt design.** Ask the candidate to write a system prompt for a specific use case in your product. This tests whether they understand how prompt structure affects model behaviour — a core AI PM skill that generalist PMs often lack. ## Mistake 3: Hiring Full-Time Before You Know What the Role Is This is the most expensive mistake. Companies hire a full-time AI PM at $180K+ per year before they have a clear picture of what the role should actually do in their specific context. Six months in, the PM is either drowning in delivery work or producing strategy documents that never get implemented. The fix is to define the role before you hire for it. What decisions does this person own? What is the interface between the PM and engineering, design, and business stakeholders? What does success look like at 30, 60, and 90 days? One practical way to define this before hiring full-time: bring in a fractional AI PM for 3–6 months. They will build the product process, define the role through actual execution, and give you a precise picture of what a full-time hire needs to look like. You end up with a much stronger job description — and often a much shorter time-to-hire. ## What Good AI PMs Actually Look Like Beyond the interview process, here are the signals that separate strong AI PM candidates from the field: **They ask about evaluation first.** Before talking about features, they ask how you measure whether an AI feature is working. This signals that they understand the fundamental challenge of AI product development. **They have shipped something with AI.** Not planned it, not advised on it — shipped it. A real product, with real users, with measurable outcomes. The more specific they can be about what worked and what did not, the better. **They can explain LLM tradeoffs in plain language.** Ask them to explain the tradeoff between a larger context window and response latency without using technical jargon. The ability to make this translation is the core of the AI PM role. **They are comfortable with uncertainty.** AI products involve more uncertainty than traditional software — model outputs are probabilistic, user expectations are hard to calibrate, and the technology is changing fast. Strong AI PMs embrace this; weak ones try to eliminate it with excessive process. If you are navigating this hiring process and want a second opinion on a candidate or a job description, [get in touch](/contact). Happy to help you get this hire right the first time.