Product-Led in the Era of AI

Still the Right Strategy, But Not the Same Strategy

Product-led growth had a good decade. Slack, Dropbox, Figma, and Notion were notable examples. The playbook was simple to describe, even if it was hard to execute. Build something so good that people try it, adopt it, and pull their whole team in behind them. Let the product do the selling. Let usage do the convincing.

For a long time, that playbook worked because the product itself was the differentiator. A better onboarding flow. A cleaner interface. A feature your competitor hadn’t shipped yet. The company that shipped faster and designed better usually won the trial.

AI is changing that equation. Not because product-led growth is dead. It isn’t. But because the thing that made it work is eroding fast, and most companies haven’t noticed yet.

Let me show you what I mean with two hypothetical company examples.

Two Companies, Two Bets

Both companies sell workflow software to mid-market operations teams. Both built their growth engine on self-serve trials, in-product upgrade prompts, and a lean sales team that only steps in once usage crosses a threshold. Five years ago, this was the entire strategy, and it worked beautifully for both of them.

Company A kept doing exactly what worked. They added AI features because customers expected AI features. A smart summarize button here. An auto-fill suggestion there. Their roadmap reads like a list of things competitors already shipped. Every feature took two sprints to build and about six weeks for a competitor to copy. Their product-qualified-lead score still fires off the same signals it did in 2021: logins, seats added, features touched. Their trial-to-paid conversion rate is quietly sliding, and nobody can explain exactly why.

Company B looked at the same market and asked a harder question. If a competitor with a foundation model API key can replicate any single feature we ship in a matter of weeks, what actually stops someone from leaving? They stopped treating AI as a feature checklist. Instead, they began treating it as an opportunity to deepen what is genuinely hard to copy: the accumulated data, the workflow context, and the trust a customer builds in the system over time. Their AI doesn’t just summarize a document. It gets better at summarizing that specific customer’s documents the longer that customer uses the product. Their Product Qualified Lead (PQL) model doesn’t just track logins anymore. It tracks whether the AI layer is making decisions the customer actually acts on.

Same starting point. Same product-led motion. Radically different trajectories, because one team understood what AI actually broke, and the other didn’t.

What AI Actually Breaks in the PLG Model

Product-led growth rested on a few assumptions that held up reasonably well for a decade. AI is quietly dismantling each one.

The first assumption: features are a moat. This was always a soft assumption, but AI has made it brittle. When any team with API access to a frontier model can bolt a “smart” feature onto their product in a sprint, feature parity stops being a competitive event and becomes a background hum. Your team ships something clever on Tuesday. Three competitors have a version of it by the following month. The rate at which good ideas get commoditized has compressed from years to weeks. If your product-led strategy depends on staying ahead on features, you are now running on a treadmill that speeds up every quarter.

The second assumption: self-serve trials show true value quickly. Product-led growth depends on a trial user reaching an “aha moment” fast, usually within the first session or two. AI features complicate this. Some AI capabilities need historical data, account context, or repeated use before they deliver real value. A summarization feature is instantly impressive in a demo. An AI feature that gets smarter with your specific data over three months is a much harder thing to prove in a five-day trial. Founders who don’t rethink their activation metrics for this reality will keep optimizing for demo-moment delight instead of durable value, and their retention numbers will tell on them eventually.

The third assumption: usage-based signals predict revenue intent. The PQL models most product-led companies built were tuned on pre-AI behavior. Logins, seats, feature clicks. AI changes what meaningful usage looks like. A single well-designed AI workflow might replace fifty manual clicks a user used to make. Your old PQL model reads that as a drop in engagement. It’s actually the opposite. Teams that don’t rebuild their qualification signals around AI-native usage patterns will keep chasing false negatives and missing their best expansion opportunities.

The fourth assumption: pricing scales with the number of seats. Seat-based pricing made sense when value scaled with headcount. AI breaks that link. A five-person team using an AI agent that does the work of fifteen doesn’t want to pay for fifteen seats. They want to pay for outcomes. Companies still pricing purely based on seats in an AI-native product are leaving money on the table with power users and pricing themselves out of smaller, high-leverage teams at the same time.

None of this means product-led growth is finished. It means the underlying assumptions require updating, and most leadership teams have yet to do that.

Why Product-Led Growth Still Matters

Here’s the part founders get wrong when they panic about AI. The instinct is to assume that if features stop being defensible, then product-led growth stops being defensible, and the answer is to pivot back toward a sales-led motion with more human intervention. That instinct is backwards.

Product-led growth was never really about the features. It was about removing friction between a customer discovering value and a customer paying for it. That principle gets more valuable in an AI world, not less. When your product’s AI capabilities need real usage data to reach their full potential, a low-friction self-serve motion is exactly how you get customers generating that data fast. When commoditization compresses the value of any single feature, the compounding advantage comes from the accumulated relationship between your product and your customer’s own data and workflows, and self-serve motion is how you accumulate that relationship at scale.

The companies that will win the next decade of product-led growth are the ones that understand the moat has moved. It used to be created by the feature set. Now it sits in three places: the proprietary data and context a product accumulates about a specific customer over time, the depth of integration into that customer’s actual workflow, and the trust a customer places in the system’s judgment after watching it perform. All three of those things get stronger with usage, which is exactly what a well-run product-led motion is built to generate. The strategy isn’t obsolete. It’s more important because now it’s the fastest way to build what actually matters.

How Companies Need to Evolve

If you’re running a product-led company, or advising one, here’s where the real work is.

Stop competing on feature velocity. Compete on data compounding. Ask your product team a different question than “what should we build next.” Ask “what are we learning about each customer that a competitor starting from zero couldn’t replicate in six months?” If the answer is nothing, you have a features business dressed up as a product-led business, and AI is about to expose that.

Rebuild your activation and PQL models around AI-native signals. The old model of counting logins and seat adds was built for a different kind of software. Go back to your data and ask what a genuinely engaged AI-native user actually looks like in your product. It’s probably fewer clicks, not more. It’s likely deeper trust signals, like a user accepting an AI recommendation without editing it, rather than raw activity counts. If your dashboards still reward raw activity, you’re optimizing your growth motion for the wrong behavior.

Redesign onboarding for delayed-value products. If your AI features genuinely improve with account-specific data over a period of weeks, your trial length and onboarding sequence need to reflect that honestly. That might mean a longer trial with a clear value curve shown to the user. It might mean seeding new accounts with representative sample data so the AI performs well from day one rather than asking a trial user to wait a month to see the good version of your product. Don’t force a fast-onboarding pattern onto a slow-value product and wonder why conversion is dropping.

Move toward outcome and consumption pricing, deliberately, not reluctantly. This is uncomfortable for finance teams that like the predictability of seat-based revenue. Do it anyway, but do it with real modeling, not a knee-jerk switch to usage-based pricing that tanks your forecast accuracy. Customers can tell when a pricing model reflects genuine value versus when it’s a defensive move copied from a competitor’s press release.

Treat trust as a product metric, not a marketing line. As AI takes on more consequential actions inside your product, whether that’s making a recommendation, taking an action, or drafting something a human used to draft, the willingness of your customer to rely on that output without double-checking it becomes one of the most important numbers in your business. Instrument it. Report it to your board with the same seriousness you report retention.

Keep the self-serve motion, but make the humans smarter about when to show up. Product-led doesn’t mean sales-free. It means sales gets involved when the product signals real intent, not on a fixed schedule. AI actually makes this easier to do well, because you now have richer usage signals to work with, provided you rebuilt the model in the step above rather than running the old one on new data.

The Bottom Line

Product-led growth isn’t a relic of the pre-AI era. It’s arguably a better fit for the AI era than it was for the one before it, because the thing AI rewards most, which is accumulated, specific, hard-to-replicate context about how a real customer works, is exactly what a good product-led motion generates as a byproduct of getting out of the customer’s way.

The companies that struggle won’t be the ones that stuck with product-led growth. They’ll be the ones that kept running the old version of it, chasing feature parity, measuring the wrong signals, and pricing like it’s 2019, while a competitor two doors down rebuilt the entire motion around what actually compounds.

If you’re a founder or CEO looking at your product-led metrics and sensing that something’s off but you can’t quite name it, that instinct is probably right. The playbook didn’t fail. It just needs a serious update, and most teams are too deep in the day-to-day to step back and do that work themselves. That’s usually the moment worth bringing in an outside set of eyes, even briefly, before the gap between you and the Company B version of your business gets too wide to close.

Product-Led in the Era of AI was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.