Are PM Interviews All About AI Now?
I keep getting asked:
What do interviewers ask these days?
How are AI skills getting tested?
Are companies only hiring AI PMs now?
So I mapped three company types across every core PM dimension. The levers are identical across all three.
Every PM, regardless of whether they work on traditional product or large language models, is ultimately answering the same questions:
- What to build — and why this, and why now?
- For whom — who is the user, what is their actual problem?
- How to measure success — what is the north star, what are the input metrics, what are the guardrails?
- How to grow — acquisition, activation, referral, expansion?
- How to retain — what keeps users coming back, what makes them leave?
- How to monetise — what is the value exchange, what is the pricing model?
This is the PM operating system. It does not change whether your product runs on an SQL database or a transformer model.
So What Actually Changed?
The application of these skills has changed. And this is where a lot of PMs get confused. Here’s a cleaner way to think about it:
In a Traditional Product : When you define “what to build,” you’re writing specs for deterministic systems. The feature does X when the user clicks Y. You can QA it. You can A/B test it. The output is predictable.
In an AI Wrapper Product : When you define “what to build,” you’re designing an experience around probabilistic outputs. The LLM might give a great answer or a mediocre one. Your job as PM shifts: you’re no longer just specifying features — you’re designing for graceful failure, setting user expectations, and deciding when AI output is “good enough” to ship.
The metrics look different too. Alongside your standard product metrics, you now track things like task completion accuracy, number of prompts required to reach a successful output, and hallucination rate. But you’re still doing the same job — measuring whether the product is delivering value.
In a Pure AI Company: Now the model IS the product. Your prioritisation decisions affect model behaviour, not just UI. Safety vs. helpfulness is a genuine product tradeoff, not an engineering concern. You need to understand human feedback training techniques, evals, and responsible scaling — not to build them yourself, but to make good decisions about them.
Get Charu Gupta’s stories in your inbox
Join Medium for free to get updates from this writer.
But even here: you’re still asking “what to build, for whom, and how do I know if it’s working?”
The Skills That Are Truly Core (and Always Will Be)
After mapping this across three company types, here’s what I believe is the non-negotiable PM core —
1. Product Strategy
2. Product Analytics
3. Product Sense
4. Cross-functional Alignment
These four are what interviewers are actually testing — even when they ask about AI. The
The “tell me about an AI product you’d build” question is a product sense question in disguise.
The “how would you measure success of an LLM feature” question is a product analytics question wearing an AI hat.
What AI Skills Do You Actually Need?
It depends on where you work — and this is the nuance that’s missing from most “AI PM” conversations.
At a Traditional Company using AI features: Understanding how recommendation systems work, how to evaluate model outputs, how to write good prompts — these make you a better PM for AI-adjacent features. But your core PM skills are the job.
At an AI Wrapper Company: You need to understand prompt engineering, context management, when RAG helps vs. hurts, and how to evaluate model quality. You’re making product decisions based on AI capability and limitations every day.
At a Pure AI Company: AI skills are table stakes. You need to deeply understand evals, fine-tuning, safety alignment, and the cost-quality-speed tradeoffs of different model configurations. You’re not building on top of AI — you’re building the AI.
The bottom line
Master the fundamentals first. Layer AI literacy on top based on where you work. The substrate changes. The operating system doesn’t.
Most companies — traditional or AI — are still testing PM fundamentals in interviews. How do you grow users? How do you retain them? How do you monetise them? AI comes up only when it’s a lever within one of these dimensions.
Yes, AI will make you more productive. Yes, knowing the basics — prompt engineering, context management, when RAG helps vs. hurts, how to evaluate model quality — will make you a better PM and help you communicate more effectively with engineers.
But unless you’re interviewing at a pure AI company, mostly nobody is quizzing you on RAG or fine-tuning. And even at AI-first companies — the first question is still: do you understand how to build a product people want?
The fundamentals are not the floor. They are the job.
