At the Berkeley AI Conference, a question came up that I've been chewing on for months: how important is a technical background for product managers compared to a background in customer research and discovery?
The answer from Perplexity's VP of Growth was the most direct version of something I've been sensing across the industry: "We're seeing a separation where engineers and designers are actually making more product decisions. We don't have these very stark roles. And so then the PM role becomes more of this: can you see the larger picture? Can you be the director of the movie and make it come to life?"
That metaphor, the PM as director, is the clearest articulation I've heard of where the role is heading. And it has massive implications for anyone building a career in product right now.
The Split
Here's what I think is actually happening, based on what I heard across four panels at Berkeley and what I'm seeing in the market:
The PM role is splitting into two distinct archetypes. The first is the technical PM who lives close to the model. At Perplexity, when they hire for AI-focused roles iterating on answer quality, they want "someone with more in-depth experience" and "data science type experience." This person is tuning prompts, designing eval frameworks, running quality experiments, and making decisions about model behavior. They're a PM in title but closer to an ML engineer in practice.
The second is the strategic PM who operates at the system level. This person owns the narrative. They see how all the pieces fit together, customer needs, business model, competitive positioning, go-to-market, organizational dynamics, and they make the calls that shape the product's direction. They're the one who decides that the home screen should be events-first instead of feed-first. They're the one who tells the CEO the keynote feature isn't ready.
Both are called "product manager." They require completely different skills.
Why This Is Happening Now
Three things I heard at Berkeley explain the split:
AI products require domain-specific technical depth. Discovery for AI features isn't the same as discovery for traditional features. You're not just figuring out what to build. You're designing for context engineering, orchestration, observability, evals, and maintenance. A PM who doesn't understand how model calls sequence, why prompt drift happens, or what an eval framework looks like can't make good decisions about an AI product. Full stop.
The speed of iteration has outpaced traditional PM workflows. Adobe's Chloe McConnell described roadmaps becoming irrelevant within months. Intercom restructured their team three times in a year. At this speed, the classic PM workflow of research, spec, handoff, build, measure, learn is too slow. Engineers and designers who are closer to the implementation are making product calls in real time because waiting for a PM review cycle would kill velocity.
Judgment is the scarce resource, not execution. YouTube's Heather Christmann said it plainly: "So much of our work now has become table stakes of what AI can do for us. The analysis, the synthesize these five points, all of that now is table stakes. And so there's an opportunity for you to really just do more critical thinking and position yourself as an owner."
The tactical PM work, writing specs, synthesizing research, building decks, is increasingly automatable. What's not automatable is the judgment layer: which bet to make, which feature to kill, when to pivot, how to frame a strategic trade-off for leadership.
What the "Director" PM Actually Does
If the PM becomes the director of the movie, what does that look like day to day?
They own the narrative, not the backlog. The director PM isn't triaging tickets. They're defining what the product is, who it's for, why it exists, and what success looks like. They're the person who can walk into a board meeting and explain why the company is betting on outcome-based pricing instead of seat-based, and have the data and conviction to back it up.
They make calls under uncertainty. Every panel at Berkeley described a world where nobody knows what the next six months look like. The director PM is comfortable saying "I don't have enough data, but here's what I believe and here's why." They use frameworks like cost-of-being-wrong analysis rather than waiting for statistical significance.
They bridge technical and commercial. Intercom's move to resolution-based pricing required someone who understood both how Finn resolves customer issues (technical) and how support leaders budget and report to their CFOs (commercial). That bridging function is the PM's core value, and it gets more valuable as the technical layer gets more complex.
They build and manage the system, not just the product. The Aakash Gupta framework from ProductCon still rings in my ears: the PMs who win aren't the ones who prompt better. They're the ones who build systems, copilots with full context, agent workflows running in the background, prototype pipelines that collapse the gap between idea and artifact. The director PM builds the machine that builds the product.
How to Position Yourself
If you're a PM figuring out which side of the split you're on, here's how I'd think about it:
If you're technical and love the model layer: Go deep. Learn evals. Understand context engineering and orchestration. Get hands-on with the tools. This path leads to AI PM roles at labs and AI-native companies where the product IS the model. Be prepared: you'll be competing with engineers who also want this role, and they have a head start on the technical side.
If you're strategic and love the systems layer: Double down on judgment. Build your track record of making good calls with incomplete data. Get fluent in AI concepts, not so you can tune a model, but so you can have credible conversations with the people who do. Ship prototypes so you can demonstrate that you can move from idea to artifact without depending on a full engineering team. This path leads to leadership roles where you're shaping product direction, not just executing it.
Either way, ship things. Michael Pratt from Apple's Platoon said the quiet part out loud: "Download Cursor, toss your resume in, and just build a resume website. That's an awesome first step of showing your product sense and your product execution." The bar for what PMs should be able to build themselves has permanently moved up.
The PM role isn't dying. It's differentiating. The generalist PM who could do a little of everything was valuable when software moved slowly and the PM was the bottleneck between research and engineering. In the AI era, that bottleneck is gone. What's left is the need for deep technical judgment on one end and deep strategic judgment on the other.
Pick your lane. Go hard.