0:00
/
0:00
Transcript

How to Product Manage AI? | Liisi German

5 Jobs of an AI PM, Buying vs Building, Becoming AI PM

Every new tool nowadays is branded "AI-powered," how do we separate the real from the buzzword? This episode digs into what actually defines an AI product - and what makes managing it so damn complex.

🎙️ Guest: Liisi German – Group Product Manager at Bolt (ex-Veriff)

From self-driving cars and document scanners to chatbots that hallucinate refunds, we talk about when AI is the right tool, and when a simple if-else statement might do the job better.

Key takeaways:

  • AI product management is about designing around uncertainty and mistakes (in some context known as hallucinations)

  • Don’t build AI unless you have to. Seriously. Start with manual hacks. Fake it, Buy it, then Build it.

  • Companies do 2 types of AI. Real AI product features and systems. And the second is automation. Most “AI” today is just automation. And that’s okay.

  • We go through examples of Bolt and Veriff, how to evaluate edge cases, label training data and more.

Framework: 5 Jobs of an AI PM

  1. Design around machine failure and edge cases

  2. Decide when to buy vs. build

  3. Define “right vs. wrong” model behavior

  4. Source, label, and manage training data

  5. Manage legal concerns

Timecodes:

0:00 Introduction

1:25 AI Product Management Overview

2:10 What is AI product?

6:13 What is difference between: AI PM and Software PM?

13:18 How to work with uncertainty in AI Teams?

24:13 How to design around AI mistakes?

28:41 Understanding Data and Training

39:30 Should you Buy or Build your AI?

56:34 What are skills of AI PMs?

1:01:11 Liisi's University course

1:05:11 Future of AI in Product Management

1:06:01 Final Thoughts and Advice

Connect:

Referenced: