AI-First Product Leadership: Hold the Paddles
If AI steers your product, the product leader must own the objective, features, data and infra.
- Difficulty
- Advanced
- Time to result
- ~months to results
- Steps
- 7
- Confidence
- 95%
Cohen's method for making product managers own AI rather than delegating it. The premise: in an algorithmic product, AI is the guide's paddle that actually steers the boat — and in most companies the person holding it is not the product leader. The framework converts 'be AI-first' from a slogan into four concrete accountabilities a PM must be able to answer on a whiteboard, plus an organizational rollout (an AI Academy, AI strategy reviews, embedded practitioners).
Origin
Developed by Tomer Cohen starting around 2016 at LinkedIn, when he took on an AI product leader role that did not exist in the company — no one had ever thought about AI from a product perspective. He institutionalized it across all product teams on becoming CPO in early 2020, consciously copying LinkedIn's 2014 company-wide mobile-first transition.
Core principles
- 01AI-first is a mindset, not a technology bolt-on. It starts in strategy, then product, then hiring.
- 02Every technological revolution dramatically changes the way we build; AI is arguably the biggest in our lifetimes.
- 03PMs treat AI as black-box magic and delegate it. That is the education gap to close, not a talent gap.
- 04An algorithm's objective is ultimately a mathematical formula — a PM should be able to write it on a board.
- 05You can build a whole strategy on data collection and fine-tuning alone, or delegate it and watch it never happen.
- 06Infrastructure is a product lever. Changing infra can beat shipping another button by a wide margin.
- 07In an AI-first product you no longer control the experience — you control the ingredients.
How to run it
- 1
Take the paddle yourself
Picture a river-rafting boat: everyone on the sides adds speed and accuracy, but the guide at the back holds the two paddles that actually steer. Those paddles are AI. Establish that the product leader — not a separate ML org — is the guide.
Pro tip The forcing question: if AI is directing your product's success and you as product leader aren't holding the paddles, what are you actually doing?
- 2
Make every PM state the algorithm's objective, in writing
For any product with an algorithm inside it, ask the owner: what is the objective of the algorithm? Can you write it down for me on a board? It is a mathematical formula and they should be able to produce it.
Pro tip Cohen explicitly challenges listeners to go ask their own algorithm-owning PMs this question and watch what happens.
Watch out If they cannot write it, they have delegated the steering of their product without knowing it.
- 3
Own the model features (parameters, not UI)
Ask what features have been added to the algorithm — meaning what parameters it learns on, not user-facing features. The PM owns this list.
Watch out Do not let 'features' be quietly reinterpreted as UI features; that is how the accountability evaporates.
- 4
Own the data collection and fine-tuning investment
Ask what investment the team has in data collection and fine-tuning. Treat this as a product strategy line item owned by the whole organization, not something the ML/engineering team is presumed to handle.
Pro tip A strategy built purely on data collection plus fine-tuning can deliver tremendous product success on its own.
Watch out Delegated fine-tuning never happens.
- 5
Go all the way down to infrastructure and inference
Push PM ownership past the model into infrastructure and inference. Improving the infrastructure underneath the product can be a far bigger lever on outcomes than building another button or surface on top.
Watch out Very few product people ever discuss the infrastructure they are building on. That is the arbitrage.
- 6
Institutionalize it: AI Academy, reviews, practitioners
Stand up mandatory AI training for every PM (LinkedIn's AI Academy, modelled on the 2014 mobile-first transition), spend your own review time on AI strategy and objectives, and seed strong AI practitioners on the product side who can teach — building distinguished leaders in waves who spread the practice.
Pro tip Spend your top personal resource — your calendar and your reviews — on the AI component in every team you enter. Attention signals priority better than mandates.
- 7
Accept that you own ingredients, not the dish
Shift from dictating the experience (this default, this onboarding order, this option set) to specifying ingredients and guidelines and letting a non-deterministic system compose the experience — with safety guardrails and responsible-AI review around it.
Pro tip Frame it as the chef who stops controlling the temperature of the broccoli and starts writing the recipe rules.
Watch out This feels scary and loss-of-control to experienced PMs. You have to genuinely believe the AI will do it better, or you will claw the control back and get the worst of both.
In the wild
When Cohen took over the feed, the AI team was centralized, outside any product team, and building toward a different purpose — what he calls a confused operation, through no fault of theirs. He unified it into a single AI-first SWAT team and personally spent most of his time on objectives, algorithm features, and data training.
→ AI became the engine of the feed's turnaround — the matchmaker that made value exchange work in both directions — and the experience Cohen credits above every specific feature. It also seeded his conviction that product people, not ML teams, must own AI.
On becoming CPO in early 2020, Cohen established an AI Academy that every PM had to pass, explicitly mirroring LinkedIn's 2014 company-wide shift to mobile-first. He spent his review time on AI strategy and objectives and placed strong AI practitioners on the product side to teach.
→ By Fall 2022, when LLMs arrived, LinkedIn's product organization was already fluent enough to restructure its entire product operations and portfolio months before the market learned about ChatGPT in March 2023.
Common mistakes
Treating AI as black-box magic and delegating it
Most product people historically thought of AI as a magic space they could not understand, so they handed it to someone else. That hands away the paddles that steer the boat.
Confusing AI-first with AI-attached
AI-first is not a technology add-on. If it does not show up in your strategy, your product decisions, and your hiring bar, you are not AI-first — you are shipping AI features.
Ignoring infrastructure as a product lever
Product leaders default to building another button or surface. Literally changing the infrastructure underneath can produce a bigger lift in product outcomes than anything you add on top.
Is it for you?
Best for
Product leaders and PMs at companies where an algorithm materially determines product outcomes, and where AI/ML currently sits in a separate org that the product team defers to.
Not ideal for
Products with no algorithmic core, or teams where the deterministic experience is a legal/safety requirement and non-deterministic composition is not acceptable.
From the transcript
“those pedals are Ai and the guy better be you”
“what is the objective of the algorithm and can you actually can you write it down for me on a board”
“what features have you added to to the algorithm and this is not user features this is specifically what parameters to learn on and then…”
“you can build a whole strategy just on uh data collection and fine tuning and your product will see tremendous success or you can delegate…”
“established an AI Academy every PM had to go through training”
“there's a realization that you don't control the experience”
“when I say I first it's not about attack it's a mindset”
From the episode
How LinkedIn became interesting: The inside story
Tomer Cohen (CPO at LinkedIn)