How LinkedIn transformed its feed into a destination product

Executive overview

LinkedIn's feed started as an activity tracker — who changed jobs, who connected to whom — and drifted into a promotional wasteland. The turnaround required changing the internal DNA of the product, not just the surface experience.

The core reframe: LinkedIn is a platform for economic opportunity built on a social graph. Knowledge exchange is one of the most powerful economic transactions. Once that premise was set, the feed had a purpose — the right experts reaching the right audiences — and everything else followed from it.

AI is the ultimate matchmaker; when it runs as a unified engine aligned to the right objective, it grows both sides of the marketplace at once.

"We might be wrong, but we are not confused"

• Clarity of thought and clarity of execution are equally important — and equally rare.
• Attaching to being right or wrong creates lingering confusion; attaching to clarity creates momentum.
• Distinguish disagreement from misunderstanding: if two people disagree, stop arguing and surface the actual point of contention.
• A decision that doesn't manifest in resourcing is not a real decision.
• The best principles have teeth — they name a trade-off, not just an aspiration.
• Confusion in a system means only luck can save you; alignment gives you a chance.

The feed turnaround: minus-one to one

• Minus-one-to-one products are harder internally than zero-to-one: entrenched metrics and processes resist change at every step.
• Set the new purpose explicitly: not a traffic springboard, not an upsell mechanism — knowledge exchange between people who matter.
• Unified the AI team under a single product objective; previously it was a separate, misaligned operation.
• Carved out 2 million members as a separate cohort to run the new experience without disrupting company-wide metrics.
• Over months, the cohort showed dramatic behaviour change — concrete evidence to bring the rest of the organisation around.
• Ran deliberate negative tests (e.g., a purely promotional feed) to prove causality, not just correlation.
• Shifted the AI objective from click-through to downstream engagement; this also surfaced and removed spam activity.

AI first as a product mindset

• AI first is a strategy and talent mindset, not a technology checklist.
• Most product leaders treat AI as a black box and delegate it — that means handing the steering paddles to someone else.
• Every PM should be able to write the algorithm's objective on a board, name the features (parameters) it trains on, and own the data collection and fine-tuning strategy.
• Infrastructure improvements can dwarf any new feature; product leaders should ask about inference and model infrastructure, not just UI.
• LinkedIn established an AI academy in 2020, required every PM to go through training, and embedded AI practitioners as internal experts.
• After the LLM wave broke (fall 2022), teams were told to drop their roadmaps, return to the problem statement, and rebuild the solution with new capabilities.

Managing the AI exploration-to-convergence cycle

• Start teams from existing objectives, not from "what can we do with this technology?"
• Allow a divergence period: let teams build competing ideas and slightly go crazy — this accelerates learning and channels the energy that would otherwise become resentment.
• Then converge top-down: pick the four or five best bets, align resourcing, and review them weekly — nothing else.
• This cycle produces a playbook (e.g., prompt engineering) that often leads the market rather than following it.

Career growth: conviction over demand

• The biggest career shift is moving from "what's most in demand?" to "what do I have conviction on?"
• Strong product leadership requires genuine excitement about what you're building; without it, impact is unlikely.
• Learn from great people continuously — many mentors don't know they are mentors.
• Product people are measured on the impact of what they built, not titles or logos.
• Identify the peak (the ambitious goal), know the base camp (a credible starting point), and accept that the middle of the mountain will be figured out along the way.

Setting ambitious goals

• Don't anchor ambition to today's numbers; start from what the product could be at its theoretical limit.
• LinkedIn's feed goal — millions of daily professionals, not tens of millions — seemed absurd at the time given where the numbers were.
• Underplaying goals to overdeliver is less useful than sharing the peak and inspiring the team to build toward it.
• With one billion members, LinkedIn still sees itself as early relative to the economic opportunity available.

Recommended books

1. Mindset — Carol Dweck: abilities are malleable; growth is always possible.
2. Thinking Fast and Slow — Daniel Kahneman: the behavioural foundation for product and organisational decisions.
3. High Output Management — Andy Grove: operational fundamentals that get a manager to a solid baseline.