I spent the first decade of my career in the recorded music industry. I started working in recording studios in 2003, went to music college, and stayed in the business through 2012. During that time, the entire industry went through a digital transformation that fundamentally rewrote how music was made, distributed, and monetized.
The recording studio, the A&R department, the label infrastructure – all of it was built around a constraint that had disappeared. Making and distributing music no longer required the apparatus that had controlled it for decades. A bedroom and a laptop could produce a professional recording. A website could distribute it to the world.
The people who survived that transition were not the ones who learned new technology the fastest. They were the ones who understood that when the cost of production drops to near-zero, the value migrates to taste, judgment, and the ability to see something that matters before everyone else does.
I’ve been thinking about that a lot lately. Because the same thing is happening to software.
The Constraint Is Moving
For as long as I’ve worked in product management, the job has been defined by a fundamental constraint: I can describe what I want the product to do, but I cannot build it myself. My output is a document – a PRD, a spec, a set of user stories – that gets handed to someone with the technical skill to translate my intent into software.
That constraint shaped the entire software development lifecycle. It created handoffs. It created wait times. It created a feedback loop measured in weeks, where the gap between what I envisioned and what got built was discovered late, when the cost of changing it was highest.
AI tools have removed that constraint for certain kinds of changes. Today, a product manager who has never written production code can take an idea for a frontend experience, express it as working software, run it against real services and real data, and share it with a colleague who can pull it up on their own machine and experience it firsthand.
What This Actually Looks Like
At Zocdoc, we’ve been living inside this shift for the past two years. Our engineering organization went through its own AI transformation – learning how to implement guardrails for AI native development, scaling our ability to deal with code that is written faster than anyone can review it, and training the entire engineering workforce on how to use these AI native tools as part of their core workflows.
Here’s what a week looks like now: A designer has an idea for a better experience on a page that our customers use daily. In the past, she would have created a mockup, presented it in a design review, and waited for it to enter a sprint. Instead, she opens the product on her machine – the real product, running locally with real services and real data – and starts iterating. She uses AI to translate her design intent into working front-end code. Within an hour, she has a branch: a version of the actual application that expresses her idea at full fidelity.
She shares the branch with her PM. She pulls it, runs it, and experiences the change firsthand – not as a screenshot in Slack, but as a live interaction on her own machine. The PM knows about our customers from speaking to them firsthand; they know what the research says about this interaction pattern and they know what a client told them last week about this exact workflow. She iterates on her branch, building on her work with additional context and then shares it back.
They go back and forth – sometimes for a day, sometimes for several – until they’ve arrived at something they both believe is right. Then they share it with their engineering partners, who review the front-end work, tighten the architecture, and scope any backend changes that are needed. This is then shared with the commercial teams, who can show a running version to a client and bring back real feedback.
Two weeks from idea to production… when it goes well. But I want to be honest about the other half of that sentence, because it is where I have learned the most. The building was the fast part. The slow part was everything that came after: the reviewing, the testing, and the less glamorous question of whether the thing was actually safe to ship.
What Stays Human
When home recording became trivially easy, a lot of people assumed recording engineers would disappear. They didn’t. What changed was the relationship between the artist and the medium. Musicians who previously needed a studio to hear their ideas could now experiment freely, iterate cheaply, and arrive at the studio – or the stage – with a much clearer sense of what they wanted. The engineer’s job shifted from translating intent to refining and hardening it. The best engineers became more valuable, not less.
The same thing is happening in software. AI tools don’t replace engineering judgment. They change when and how that judgment gets applied. When a PM or designer can produce a working front-end draft, engineers spend less time translating specs and more time on the work that actually requires their expertise: performance, scalability, security, production hardening. The draft arrives closer to right. The feedback happens earlier. The quality goes up.
This shift changes what it means to be a PM. When I can see the product as the customer sees it – with all its feature flags, A/B tests, and configuration variations – I form better hypotheses. When I can iterate on those hypotheses in code rather than in a document, I test them faster. When I can share a running version of my idea rather than a description of it, the conversation with stakeholders becomes more honest.
The judgment, the taste, the understanding of the customer – those are still mine. The tool just lets me express them directly at a speed that makes real iteration possible within a single sprint.
The Bigger Bet
I voluntarily left a role leading a team of PMs to become the first test case for this model. I call myself the canary in the coal mine; if a PM can become a builder, ship production code, and create real customer value without breaking quality or safety, then the model works. If I can’t, we learn that too, and we learn it cheaply.
I believe we are the last generation of people who will enter the workforce as “product managers” and “product designers” in the way those titles have been understood for the past twenty years. The roles won’t disappear. But what it looks like is changing as fundamentally as what it meant to be a musician changed when the studio moved into the laptop.