AI isn’t coming on your job. It’s coming on your ambiguity.
That was one of many greatest takeaways from our dialog with Digvijay Singh, Senior Product Supervisor at YouTube Advertisements, who’s spent years working on the intersection of machine studying, information science, and product technique. In our newest episode of Productside Tales, Digvijay unpacked the approach AI Product Administration roles are consistently shifting — and what it actually takes to achieve every layer of this fast-shifting stack.
For seasoned product leaders or bold PMs who wish to talent up your AI fluency, right here’s what you could know.
The Three Layers of AI Product Administration
When most PMs take into consideration AI, they image flashy purposes: copilots that write code, chatbots that promote, or advice engines that predict the subsequent binge. However Digvijay breaks the ecosystem down into three distinct layers, every with its personal product mandate, talent set, and enterprise impression:
- Infrastructure Layer: These are the PMs constructing the foundations: information facilities, GPUs, cloud clusters, and the pipelines that feed mannequin coaching.
- Mannequin Layer: These PMs outline and refine large-scale fashions themselves (from LLMs to diffusion and retrieval fashions) setting the metrics that make intelligence measurable.
- Software Layer: That is the place most product managers reside at this time. It’s about embedding these fashions into merchandise that ship actual consumer worth — from advert optimization to artistic technology.
Every layer requires a special product mindset. The Infra PM thinks in capability, price, and latency. The Mannequin PM thinks in evals, fine-tuning, and step-change innovation. The Software PM interprets all of it into buyer outcomes.
As these AI Product Administration roles mature, PMs will more and more specialize by layer, whereas leaders who perceive all the stack will develop into the brand new strategic connectors inside their orgs.
From Roadmaps to Evals: The New Foreign money of Success
Conventional product managers reside and die by OKRs. However in AI, success seems completely different.
“You don’t outline success by OKRs anymore,” Digvijay explains. “You outline it by evals.”
In different phrases, you’re not simply measuring product outcomes. You’re coaching the system to acknowledge what good seems like. Evals (quick for evaluations) are the new success metrics of AI Product Administration, defining the bottom reality your mannequin learns from.
Think about you’re optimizing advert experiences on YouTube. Your evals would possibly measure when a consumer chooses to skip an advert, how lengthy they interact, or whether or not the advert matches contextual intent. These metrics don’t simply let you know efficiency — they train the mannequin what to do subsequent.
For product leaders, that’s a profound shift. Your “necessities doc” now doubles as a coaching framework for the product’s intelligence. You’re not simply writing consumer tales; you’re writing instruction units for studying programs.
This mindset marks a significant evolution in AI Product Administration roles, the place PMs at the moment are accountable not only for outcomes—however for shaping how machines be taught to ship these outcomes.
Each PM Is Now an AI PM. However Not Each PM Is aware of It
There’s a well-liked perception that “each product supervisor is now an AI product supervisor.” Digvijay agrees, however there’s a twist.
Sure, AI will contact each product floor finally. However true AI product administration calls for a brand new layer of instinct: understanding how fashions suppose, fail, and enhance.
That doesn’t imply you could code or construct neural nets. It means you could know sufficient about AI mannequin conduct, information dependencies, and analysis methods to information cross-functional selections with credibility.
The perfect AI product managers don’t ask, “Can we add AI to this?” they ask, “The place does intelligence really create worth?”
That distinction separates the resume titles from the operators.
Essentially the most impactful AI Product Administration roles will belong to PMs who can reply that query with readability, anchoring their technique in enterprise worth.
The Rise of the Mannequin-Layer PM
Whereas most PMs at this time work within the software layer, Digvijay predicts a surge in demand for model-layer PMs: those defining how LLMs, multimodal fashions, and agentic programs evolve.
Why? As a result of that’s the place the step-change innovation is occurring.
“Mannequin-layer PMs get one of the best of each worlds,” he says. “They’ll leverage all of the infrastructure that’s being constructed and nonetheless invent new purposes that didn’t exist even a yr in the past.”
These roles demand deep cross-disciplinary fluency (understanding information pipelines, analysis frameworks, and moral issues) and a willingness to suppose from zero to 1.
If infrastructure PMs scale capability and software PMs scale adoption, model-layer PMs scale risk itself.
AI Abilities for Product Managers: What to Grasp Subsequent
So, what does it take to thrive throughout these AI product administration roles? Based on Digvijay, there are two must-have abilities each PM ought to make investments on this yr:
- Constructing Product Instinct with AI Instruments
– Spend time with the merchandise shaping the sphere — from Claude and Gemini to Midjourney and Synthesia.
– Don’t simply use them; reverse-engineer how they work. Ask: what’s the mannequin optimizing for? What information patterns is it studying from?
– This hands-on curiosity is the way you’ll sharpen your intuition for what’s doable (and what’s hype). - Writing Evals Like a Trainer, Not a Tester
– Consider the mannequin as your intern: it’s important to prepare it on what good seems like.
– Each eval you write is a studying scaffold. Readability, context, and edge-case pondering matter as a lot because the product spec itself.
Collectively, these abilities elevate PMs from “AI customers” to AI architects: product leaders who can form how intelligence itself will get constructed.
The Subsequent Frontier: Product Technique at AI Pace
Digvijay predicts that the market will quickly worth PMs who can function throughout a number of layers: fluent sufficient in infra, mannequin, and software to attach the dots between technical functionality and industrial technique.
In different phrases: the brand new 10× PM isn’t the one who ships quicker. It’s the one who decides smarter.
That’s the way forward for product management within the AI period. not managing roadmaps, however managing reasoning.
Lead the Shift. Degree Up Your AI Product Administration Roles
- Hearken to the total dialog. Catch Digvijay Singh and Rina Alexin on Productside Tales — accessible now on Spotify, Apple Podcasts, and Amazon Music. You can even watch it on YouTube for the whole interview.
- Take your understanding of AI PM roles even additional with our AI Product Administration Certification.
- Now it’s your flip. How are you evolving your AI Product Administration roles into 2026? What new abilities or frameworks are serving to your groups bridge technique, information, and intelligence? Share your ideas and tag Productside on LinkedIn — we’d love to listen to the way you’re main the subsequent wave of product innovation by AI.
