The Three Flavors of AI Product Management
Map any AI PM role into one of three types to target the right skills and job.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 3
- Confidence
- 85%
A taxonomy that splits 'AI PM' into three distinct roles: platform PMs who build tooling for AI engineers, AI-product PMs who package a core model into a consumable experience, and AI-powered PMs who use AI to build any product more efficiently. Knowing which flavor you want clarifies what to learn, what portfolio to build, and which companies to target. Khan predicts most PMs will soon be one of these three as AI becomes 'as common as the database.'
Origin
Aman Khan's own framing, developed as Director of Product at Arize AI (an AI observability/evaluation platform) and across prior roles at Spotify's ML platform team, Cruise, Zipline, and Apple.
Core principles
- 01'AI PM' is not one job — the required depth and skills differ sharply by flavor
- 02Platform PMs push technology forward; product PMs package researchers' technology; AI-powered PMs apply existing models to their own product
- 03You rarely need to build a core model from scratch — most value is in application
- 04AI is trending toward being infrastructure, like a database, embedded in most SaaS
How to run it
- 1
Identify the platform PM path
This role builds tools for AI engineers — observability, evaluation, and the clunky new interfaces for building on top of LLM providers. You are responsible for pushing the technology forward for a technical user base.
Pro tip Suited to people who get obsessed with hard, deep-in-the-stack technical problems that only reveal their complexity as you dig in.
- 2
Identify the AI-product PM path
Here the core product is centered on AI — the model under the hood is the secret sauce (e.g. ChatGPT, NotebookLM). Your job is to package researchers' and engineers' technology and make it consumable for a business, enterprise, or consumer.
Watch out Historically this path demanded deep ML background — training data, data splits, launch infrastructure — so weigh whether you want that depth.
- 3
Identify the AI-powered PM path
You are not building the core model from the ground up; you use large language models (or any models) to build the best experience for your customers and to be more productive across the whole PM role.
Pro tip This is the most accessible on-ramp today and where most PMs will trend as AI becomes pervasive.
In the wild
Arize is an observability and evaluation platform that started with machine learning models (ranking, regression, classification) and has been pivoting to broadly any AI, increasingly large language models. The tooling to answer 'is my app doing the thing I expect?' is very early because people have only built on LLMs for a couple of years.
→ Khan positions himself as a platform PM enabling AI engineers to understand model impact, illustrating the first flavor in practice.
Khan cites ChatGPT and NotebookLM as products where the core experience is enabled by the model underneath — researchers push the boundary and the PM packages that technology for consumption.
→ Serves as a concrete anchor for the second flavor, distinguishing model-centric products from AI-assisted ones.
Common mistakes
Treating 'AI PM' as a single monolithic job
Lumping all AI PM roles together leads to preparing the wrong skills — e.g. grinding deep ML theory for a role that only needs applied prototyping, or under-preparing for a model-centric role that demands real ML depth.
Assuming you must build the core model
Many aspiring AI PMs think they need research-team resources to compete. Most opportunity is in the AI-powered and product flavors, where you apply existing models — believing otherwise keeps people from even starting.
Is it for you?
Best for
Product managers deciding how to break into or specialize within AI who need to target their learning and job search
Not ideal for
Teams that already have a clear, narrow AI mandate and don't need a career-orientation map
From the transcript
“I really break it down into maybe three flavors of AI management so the first is AI uh platform PMs and that's kind of where…”
“AI product PMS are really the flavor of PM where the core product is centered around Ai and an example of that would actually be…”
“really what what an AI powered product manager in my mind does is you're not really responsible for building the core model from the ground…”
“it almost feels a little bit like AI will be as common as the database to some degree for SAS applications”
From the episode
Becoming an AI PM
Aman Khan (Arize AI, ex-Spotify, Apple, Cruise)