OpenAI's Framework for AI Product Sense (Used by Top PMs)

5,500 words of battle-tested principles: From probabilistic fit to system design, learn what separates $1.2M failures from products users trust

Product sense is one of the haziest topics in product management:

• What does it refer to, exactly?

• Is it just bullshit?

Many of us are left wondering what this mythical product sense is. This is especially true when it comes to AI Product Sense. Is it even real?

---

My Long-Time Thought Partner on AI PM

I decided to wrestle with these thoughts with someone at the center of AI product development, Miqdad Jaffer. As a PM leader at OpenAI, he’s lived AI product sense.

P.S. The next cohort of Miqdad’s AI PM certificate (which I took in 2024) starts October 18th:

---

Today’s Deep Dive

We’re unpacking AI Product Sense step-by-step to a level no one has even done before:

1. Defining Product Sense

2. Defining AI Product Sense

3. The Key Lenses of AI Product Sense

4. 10 Core Principles to Improving Your AI Product Sense

5. System Design AI Product Sense

6. GTM AI Product Sense

7. Improvement Rituals

---

1. Defining Product Sense

Two of the best explanations of product sense I’ve seen are from Sid Arora and Shreyas Doshi.

Let’s take a look at each.

Here’s Sid’s:

Sid defines product sense as the ability to take a vague and ambiguous problem statement, take it through user, problem, and solution discovery, deploy the solution, and measure results.

Now let’s look at Shreyas’ definition:

Shreyas doesn’t per se define it. He identifies levers of improving your product sense as: motivation theory, cognitive empathy, creativity, domain expertise, and communication.

What you can see from both of these is that they are very broad definitions both of what product sense is, and how you can improve it.

Product sense is not a small topic. It’s basically all of the product thinking a PM needs to do.

We can formally define it as:

Product sense is the ability to make consistently good product decisions; knowing what to build, for whom, and why. It’s pattern recognition across user needs, market dynamics, and execution tradeoffs

Now, let’s move on to AI product sense…

---

2. Defining AI Product Sense

When I was a VP of Product at Apollo.io ($2.5B Sales Tech SaaS), we built an AI email writer.

The demo was perfect. Our CEO, Tim, who had a notoriously high bar, actually loved it. The model responses were great. As were the interface and marketing material.

The sales team jumped at selling it.

Three weeks later, it was dead.

Users sometimes loved the emails. Other times, it destroyed their spam rating and tanked their inbox. Trust was being evaporated. And it was costing us an arm and a leg.

So we paused that version of the feature. $1.2M of development seemingly wasted (3 months of the 5 people building it).

That was when I learned the invisible line that separates product sense and AI product sense. Traditional product sense failed us in the AI scenario:

1. The models in AI were non-deterministic

2. And the costs weren’t fixed

These distinctions make AI product sense its own skill, unique from product sense.

So how do we define AI product sense? In plain english:

More formally:

AI Product Sense is the fit between need, model behavior, and the economic envelope that lets that behavior repeatedly deliver value under uncertainty.

Here’s what I mean.

At Apollo, when we came back and tweaked everything:

• The underlying context engineering

• The economics of the models we were using

• The feedback loop that we were getting from users

We ended up shipping the highly successful email writer you see in the product today (which drove millions in revenue).

You need to have a highly tuned AI product sense to avoid those wasted development cycles, and ship features that land the first time.

As Marty Cagan says, it’s something you can and need to improve.

So how do you build this AI product sense? Miqdad and I have put together the deepest guide ever developed (5.5K hand-written words, without any LLM-fluff).

---