Skip to main content

The AI-Informed Product Manager

5 min read
Tpm

Tpm

You don't need to code. You need to know what AI can and can't do. That's product judgment.

The AI-Informed Product Manager

TL;DR

  • PMs don't need to be AI experts. They need to know enough to scope, prioritize, and communicate with eng.
  • "AI-native" products are just products. User needs first. AI is a means, not the end.
  • Your edge: understanding both the problem space and the solution space. AI changes the solution space. Stay curious.

The best PMs in 2026 aren't the ones who can train a model. They're the ones who can articulate what AI should and shouldn't do in their product, work with engineers who do build it, and keep the user at the center when everyone else is chasing the tech.

What "AI-Informed" Means

Enough technical literacy to have a conversation:

  • You don't need to know how transformers work. You need to know: What can current models do well? What do they hallucinate? What's cheap vs. expensive (latency, cost)?
  • Read one or two solid explainers. Ask your eng team to translate the rest. Stay curious.

Understanding AI as a capability, not a feature:

  • "We're adding AI" is not a product strategy. "We're using AI to do X for the user" is.
  • AI can summarize, generate, recommend, classify. It can't reason reliably, guarantee correctness, or replace human judgment in high-stakes contexts. Design for that.

Scoping and de-risking:

  • "Can we ship a v1 with AI doing X?" You need to know if X is in the "AI does this well" bucket or the "AI is inconsistent here" bucket.
  • When in doubt, prototype. Don't over-commit in a PRD to AI capabilities you haven't validated.

How to Work With AI-Building Teams

Speak their language (a little):

  • "Fine-tuning" vs. "prompting" vs. "RAG" — rough idea. You don't need depth. You need enough to understand timelines and trade-offs.
  • "How long to add X?" — If they say "it's a prompt change" vs. "we need to retrain," that changes your roadmap. Know enough to ask.

Protect the user:

  • Engineers will optimize for model performance. You optimize for user outcome. Sometimes those align. Sometimes they don't.
  • "The model is 95% accurate" — okay. What happens to the 5%? That's a product question. You own it.

Scope for failure modes:

  • AI will hallucinate, go off-rails, or break edge cases. Your spec should address: What do we show when we're not confident? What's the fallback? Who does the user talk to when it goes wrong?

What Makes You Indispensable

  • Prioritization. AI can suggest. You decide what we build and in what order. That's strategy.
  • Stakeholder alignment. AI doesn't sit in the meeting with Sales, Legal, and Support. You do.
  • User empathy. AI doesn't watch someone struggle with your product. You do (or should).
  • Judgment under ambiguity. When the data is fuzzy and the options are many, you make the call. AI provides options; you choose.

PM: vague on AI. Cannot scope or communicate with eng. AI as black box.

Click "AI-Informed PM" to see the difference →

Quick Check

What does an AI-informed PM need to know?

Do This Next

  1. Have one "AI 101" conversation with an engineer — Ask: "What can our stack do today? What's the gap?" Take notes. Refer back when scoping.
  2. Audit one product area — Where could AI help? Where would it hurt? Write a one-pager. Share with eng.
  3. Add "AI literacy" to your learning queue — One article or video per week for a month. You don't need a PhD. You need enough to contribute to the conversation.