They Tried to Kill Product Managers. Now Everyone Needs to Be One.
AI makes building easier. Figuring out what to build and getting it into the right hands? That's as hard as it's ever been. Welcome to the job you didn't know you had.
Remember when people declared product management was dead? That was cute.
The moment that crystallized it: Figma’s Config 2023. Brian Chesky told a crowd of designers Airbnb had “gotten rid of the classic product management function.” The audience cheered. One tweet on the topic got six million views. Designers celebrated. PMs panicked.
Chesky later clarified he’d merged product management with product marketing. But the sentiment had spread. The broader critique wasn’t entirely wrong. A lot of PMs had spent years becoming backlog administrators and Jira ticket managers. That’s project management, not product management, and people could see it.
Then AI piled on. If everyone can build, why does a specialized role exist to manage the process?
The “PM is dead” crowd must feel pretty vindicated right now.
Except they’ve gotten it almost completely backwards. AI isn’t killing the need for product management. It’s making product management a skill everyone needs.
Building Got Easier. The Hard Parts Didn’t.
Yes, AI has compressed the time and complexity of actually making things. A solo founder can build a working web app in a weekend. A marketer can spin up a prototype without filing a ticket. Anyone with an idea and some patience can ship something real.
I’ve been living this myself. I’m building stuff like crazy and it’s genuinely fun. Over the past few months I’ve built an LLM/SEO optimizer, an innovation toolkit, an applicant tracking system, an investor portal, a venture studio readiness tool, and a pile of automations. I’m a product person, not a developer. The line is getting blurry in ways I didn’t expect.
So yes. Building got easier.
But figuring out what to build? Still incredibly hard. Getting it into the hands of the right people and making them actually care? Harder than ever.
The cost of building wrong hasn’t gone down just because building got cheaper. If anything, it’s gone up. Because now everyone is building. The market is flooding with products, which raises the bar for everyone. Differentiation is harder. Getting your message out is harder.
And you know who else is building, or at least thinking about it? Your customer. If they look at your product and think, “I can just build that myself,” you’re in trouble Even if they can’t, the perception is powerful enough to kill a sale.
You’re Already Doing Product Management. You Just Don’t Know It.
If you’ve written a single prompt into an AI tool to build something you’re doing product management.
Every time you decide what to build next, you’re doing product management.
Every time you talk to a user (or skip talking to a user), you’re doing product management.
Every time you choose which features to include, which to cut, and what success actually looks like, you’re doing product management.
Every time you think about how someone is going to find what you built, you’re doing product management.
You’re doing it whether you have the title or not. Whether you’ve read a book or blog post about it, or listened to a podcast.
The question is whether you’re doing it well.
What I find fascinating is that product management skills are the very skills you need to use AI effectively to build something.
You need to think clearly about problems. Articulate what you want with precision. Break complex work into pieces that can be delegated. Synthesize messy, incomplete information into a decision. Catch when AI is confidently wrong and know how to course correct.
That’s not just AI literacy. That’s product management.
The moment everyone can build, the skills that were supposedly obsolete become the skills that determine whether you build something worth using.
So What Are These Skills, Actually?
Product management confuses people because everyone has a different definition. The core is simple: find a real problem worth solving, build the right-sized thing that solves it, get it into the hands of the people who need it, and learn constantly throughout. It’s not easy, but it always felt pretty clear to me.
Here’s what it looks like in practice:
Know whose problem you’re solving. Not who you think has the problem. Who actually has it, how badly, and what they’re doing about it today. If you can’t answer those questions specifically, you’re building on assumptions. Assumptions that feel true are still assumptions.
Define success before you build, not after. What specific behavior change in users tells you this is working? What does the number look like in 90 days if you’re right? Set it before you write a single prompt. Otherwise you’ll rationalize whatever you ship as working, because you want it to be working.
Scope is a decision, not a default. AI makes it trivial to add features. That’s a trap. Lenny Rachitsky defines the job as “delivering business impact by marshaling your team to identify and solve the most impactful customer problems.” Not shipping everything you can think of. The best products do fewer things exceptionally well.
Distribution isn’t someone else’s job. Figuring out what to build is hard. Getting people to use it is equally hard. Product managers aren’t growth bystanders. Neither are you. If you’re building without thinking about how someone discovers, tries, and keeps using what you made, you’re building in a vacuum.
Your product needs to feel right, not just work. Use your own product obsessively. Feel the friction as a genuine first-time user. Notice what people don’t say in feedback sessions but reveal in their behavior. The judgment you build from this, over time, is what makes the difference between something that’s functional and something people actually want.
None of this is revolutionary. All of it gets skipped by people who just want to build.
Learn It By Doing. Not By Outsourcing It.
AI has a way of tempting us into being…brain dead. It’s so easy to agree with everything it outputs and stop reading and thinking.
AI can do a remarkable amount of PM work. It can synthesize user research, spot patterns in feedback, draft requirements, flag edge cases, simulate user reactions. That capability is real and improving fast.
But if you’ve never done these things yourself, you can’t tell when AI is wrong.
The judgment required to evaluate what AI gives you has to come from somewhere. It comes from talking to users, watching them use your product, shipping things that fail, understanding why they failed, and trying again. From the customer call where the data said one thing and the room said another, and knowing which one to trust. That instinct doesn’t come from reading about it. It doesn’t come from prompting your way through it. It comes from doing it.
Shreyas Doshi puts it plainly: “The only real long-term career moat for product people is how you can improve on the already-brilliant, already-comprehensive inputs and outputs that AI will provide for you.”
You can’t improve on something you’ve never tried to do yourself.
So if you’re new to building, here’s where to actually start:
Talk to five people before you write a single prompt. Not friends. Find five strangers who actually have the problem you think you’re solving. Understand how they solve it today, what it costs them, whether they’d change their behavior for your solution. If you can’t find five people with the problem, you probably don’t have a product yet.
Write your first product requirements by hand. Type it out. No AI help on this one. Describe what you’re building, who it’s for, what problem it solves, what success looks like, and what the riskiest assumption is. Then ask AI to evaluate it. You’ll learn more from writing it yourself than from anything AI could produce for you.
Launch smoke tests to validate the value proposition and your ability to get anyone to the front door. A smoke test is a way to see if you can attract users, but there’s nothing real yet. You might have a landing page with a few ads. Maybe you share some things on social media. You’re trying to test interest, perhaps a bit of conversion (e.g., through email signups). A lot of people skip this because they build and launch. When it doesn’t work, they’re not sure why, and there’s been little learning.
Build the smallest thing that tests your core assumption. Not the full vision. The one feature or flow that proves whether your hypothesis is true. Everything else is scope creep. AI will make adding more feel painless. It’s not. Don’t know what your core assumption is? Stop. Go back. That’s the first thing to figure out.
Use your own product as a stranger every week. Pretend you’ve never seen it. Where are you confused? What’s missing? What feels broken? This is not optional and it never gets easier, because your familiarity with your own product degrades your ability to see it clearly.
Expect to be wrong in important ways. Not about everything. But about the things that matter most. That’s how this works. The goal of a first version isn’t to nail it. It’s to learn what you got wrong fast enough to fix it before you’ve built too much on top of it.
The faster you can build, the more important it is to have done the thinking first.
One Warning About AI and Product Skills
There’s a growing category of AI tools specifically designed to accelerate product management work. Frameworks, prompt libraries, agentic workflows that cover everything from discovery to strategy to go-to-market. A good example is pm-skills on GitHub, built by Pawel Haruyn: 100+ agentic skills replicating PM workflows end to end, nearly 9,000 stars. It’s genuinely impressive work and worth knowing about.
I use tools like this myself, including ones we’ve built at Highline Beta. In the hands of someone who knows what they’re doing, they’re a real accelerant.
That last part is the key.
These tools do a lot of things quickly, but without the product management fundamentals in your back pocket and an instinct for what’s good or bad, you’re following blindly.
This is the same failure mode PMs fell into before AI, just wearing different clothes. Before, it was mistaking status checks and coordination for product work. Now it’s running AI workflows and mistaking output for thinking.
The skills aren’t in the tool. The tool accelerates the skills. You have to build those skills first, or the acceleration is just moving faster in the wrong direction. (To be clear, I’m not referring to Claude Skills in this paragraph. 😆)
Product Management Isn’t Dead. It Became Everyone’s Job.
The irony of the “PM is dead” moment is that it was half right. The version of product management that died deserved to die: the backlog-managing, sprint-facilitating, paper-pushing version that confused coordination with creation.
It was never about that. It was always about judgment. About figuring out what’s worth building, for whom, and why. About caring enough about the user to understand them, and caring enough about the business to connect those two things.
That work doesn’t go away because building got easier. If anything, it matters more. When everyone can build, the products that win are the ones built by people who understand what they’re actually trying to do.
You’re going to build something with AI. You probably already are.
Guess what? You’re a product manager now. The only question is what kind.
What’s the hardest part of building with AI that no one prepared you for? Drop it in the comments.
These resources may help you get started:
The line that stuck with me is that the skills aren't in the tool, the tool accelerates the skills. You have to build the judgment first, or the acceleration just moves you faster in the wrong direction. That's true for product management and pretty much everything else AI touches.