AI doesn’t just change what we build. It changes what it means to be excellent at building.
For years, PM “output” has been confused with PM “value.” We rewarded the person who could produce the most detailed spec, the cleanest deck, the most comprehensive set of edge cases. In an AI-native world, that work is table stakes. A competent LLM can draft a thorough PRD in five seconds. It can translate meeting notes into tickets. It can generate five roadmap narratives in the time it takes you to refill your coffee.
So if the artifacts get commoditized, what becomes scarce?
The job is no longer to be the human compiler that converts ambiguity into documents. The job is to be the human compass that converts ambiguity into direction.
AI makes it easy to generate “a plan.” It does not make it easy to generate the right plan. The PM advantage moves up the stack:
This shift stopped being theoretical for me the day I started seriously playing with coding agents.
Over the last few months, I’ve bounced between tools like Cursor, Lovable, Kiro, and Claude Code, mostly out of curiosity at first. I expected incremental productivity gains. What I got instead was a redefinition of what “work” even means. Tasks that would normally take a team weeks or months, scaffolding a new app, wiring up flows, refactoring a messy module, generating tests, suddenly collapsed into minutes.
It was exhilarating and slightly unsettling. Not because it was “perfect,” it wasn’t, but because it was close enough to create momentum instantly. The bottleneck moved away from implementation speed and landed squarely on something else: my ability to describe what I wanted.
Here’s the humbling part: the hard problem wasn’t getting the agent to code. The hard problem was translating a vivid picture in my head into prompts that are crisp enough for an agent to execute. When the agent went off track, it was rarely because it was “dumb.” It was because my intent was fuzzy, my constraints were implicit, or my definition of “good” lived in my intuition instead of in words.
That experience is the future of PM work in microcosm.
If AI can generate the spec instantly, it exposes what the spec was really doing. Often it was compensating for a lack of clarity: unclear strategy, fuzzy success metrics, unresolved tradeoffs, or misaligned stakeholders. A longer doc was sometimes a substitute for a sharper decision.
In the new model, the PM’s contribution is not “more detail.” It’s:
An LLM can draft words. It can’t own consequences.
AI-era PMs will increasingly look like three things:
Curator:
You are surrounded by options. Your edge is selecting what’s coherent and what’s worth it. You curate not just features, but experiences, behaviors, and tradeoffs.
Conductor:
You don’t just “manage a backlog.” You orchestrate a set of humans and agents. You decide when to delegate to AI, when to pull humans in, and how to blend them into a workflow that feels smooth rather than chaotic.
Culture builder:
When AI accelerates everything, the temptation is to become transactional. But the highest-performing teams are not the ones who move fastest for a week. They’re the ones who sustain intensity for a year. Warmth, psychological safety, and genuine relationships are competitive advantages. They reduce friction. They increase candor. They make conflict productive instead of corrosive.
If I had to summarize the shift: PM moves from writing to judgment. And judgment is built from:
That’s the future. The PM who wins is not the one who writes the longest spec. It’s the one who can face a messy, high-stakes problem and crisply define the goal, the why, the quality bar, the fastest learning loop, and the human trust we must protect.
In the age of instant text, being deeply human becomes the differentiator.