The Old Playbook That Built SaaS
When I think about my career building products at Microsoft, Slack, and Spendesk, etc. I realize how much of the products’ success was tied to the Product-Led Growth (PLG) playbook of the 2010s and early 2020s.
At Microsoft, we scaled Outlook mobile from 1M to 60M MAU through relentless friction reduction, usage loops, and enterprise readiness. At Slack, my teams pushed beyond messaging into platform growth, architecting integrations that made the product indispensable, and leveraging email to drive the funnel. We built the fastest-growing PLG machine in the fastest-growing PLG company.
These were category-defining wins, and they were made possible by a PLG system built on assumptions that users:
Were willing to invest time learning your product.
Wanted to explore value progressively.
Would tolerate onboarding journeys to get going quicker.
Measured you against competitors in the same category.
Those assumptions no longer hold.
AI has changed the game. Today’s users don’t want progressive disclosure. They don’t want to learn your UI. They don’t want onboarding tours (they never really did!). They don’t even want to sign up if another tool (or ChatGPT itself) can solve the problem in seconds.
The PLG era we built SaaS in is over. But PLG isn’t dead. It’s being rewritten.
In this piece, I’ll outline four shifts every founder and product leader must grapple with. And in doing so, I’ll connect the old playbook, the one that worked for me and countless other leaders, with the new realities of the AI-native era.
Time-to-Value Expectations Have Collapsed
In the old world of PLG, the magic came from progressive disclosure. You revealed value step by step, pulling users deeper into your product over days. A free trial, a guided tour, maybe some nudges and lifecycle emails. These worked because users were willing to spend the time. It was what users were used to and it was better than what had come before. Products like Photoshop or Microsoft Word - powerful tools, but with every feature imaginable vomited onto the screen for the user to pick through.
At Outlook mobile we leaned into the concept of progressive disclosure. Once we realized the mobile experience was about consumption and not creation, we found ways to remove features from the top level of the information architecture. This was a fight, because customers and folks in Redmond expected Outlook mobile to look and behave like Outlook on Windows. This was why Microsoft had lost in the past. And going against this with Outlook mobile was a key reason why we won. Against Gmail and Apple mail no less.
But that was 2016. Today, it would be too slow. More importantly, it was a fixed approach. We were making decisions that would improve the experience and drive engagement and growth. But it wasn’t adaptive.
There hasn’t been a massive amount of research in this domain, but we’re already seeing user expectations moving in this direction. And as with many growth philosophies and approaches, it’s coming from the domain of eCommerce and sales. According to McKinsey, 71% of customers expect companies to understand their needs and personalize their experience. According to Comviva, customer acquisition and retention costs can be reduced by 28% as a result of personalization.
Why does this expectation translate? Because AI has trained them. When ChatGPT, Perplexity, or Claude can generate an answer in three seconds tailored to your query, every second of lag in a SaaS product feels like friction. Expectations have been reset, and the bar is high.
In the AI era, time-to-value isn’t a funnel metric. It’s a second-by-second expectation.
Join me at the Product-Led Summit, where I’ll be presenting on this topic!
What This Means for PLG Today
Freemium alone is no longer enough. Users won’t wait days to find value. The “aha” moment must happen in seconds, not sessions. Immediately upon sign-up, the user expects to be receiving value. The expectation is that within those few seconds the user can accomplish what they had in mind the moment when they clicked your sign-up button.
AI-powered personalization is mandatory. The single prompt input box of your favorite GenAI tool is just the starting point. While there is customization in the text response, there is no personalization in the visual experience. Just look at how ChatGPTs UI quickly becomes a disaster once you have more than five projects running in parallel. The point is, your product must infer the context of the user and then adapt to it. What is their role? What is their real-world workflow? What is their goal and objective? Your product must do that before the user clicks anything. And then adapt to meet those inferred expectations.
Metrics shift. There needs to be a change in what we measure, how we measure it, and what does success look like. Back at Slack in 2021, success was defined by the number of messages sent within five days by how many people within the channel - days. Now we need to measure the accomplishment of perceived value in how many seconds. Instead of “days to first value,” we measure “seconds to perceived value.”
The companies that win won’t just remove friction. They’ll erase it.
The Learning Curve is Dead
Classic PLG assumed users were willing to invest time learning your product. Tooltips, help docs, and sandbox accounts were all part of the journey. The early masters of this was Evernote. Remember them? Once upon a time, they were the productivity innovation darlings of Silicon Valley. They developed a novel approach to note taking based upon a deep understanding of how the user thinks about taking notes and how they actually use them after the fact. This novelty introduced complexity to the user and therefore friction to the creation of habits which impacted retention and growth.
Rather than hide all of the complexity (as Notion originally did), they provided in-app tool tips and educational moments delivered just at the right time. It was a blessing and a curse. Everyone fell back on this affordance to build engagement and drive retention at the expense of thoughtful design. But it worked.
At Microsoft, we obsessively reduced friction in Outlook mobile. We simplified setup, built intuitive gestures, and optimized the core mail–calendar integration. That focus was key to differentiating and scaling from 1M to 60M MAU.
We always had “educationals” on our backlog, but we kept on deprioritizing it. Emphasizing the “less is more” philosophy. A good example was the removal of the files tab. The outrage! At Slack we took a hybrid approach as we were better resourced both in design and engineering. Prompts were introduced, and you see them today. Nudges to move people along. We were also constantly paring back on design. Keeping the visual noise to an absolute minimum, following what I call the “successive revelation” approach to the user experience.
But here’s the truth: even the level of friction we considered a triumph then would kill adoption today. AI has flipped the relationship. Products can no longer expect users to learn them. Products must learn the user.
The numbers back this up. According to MarketsandMarkets, the conversational AI market is projected to grow at a 19.6% CAGR, from $17.05B in 2025 to $49.8B in 2031. Why? Because users don’t want menus, inputs, or training. They want to describe what they need and have the system respond.
But this only tells part of the story. It’s directional, but shouldn’t be taken literally. In that timeframe, software will move from reactive to responsive to proactive. Even now, ChatGPT and Claude are learning what you want in reaction to what you input. You’re not clicking a single mouse button on a menu, you’re tapping on hundreds of keys of the keyboard. You’re telling the AI what you want and slowly the system starts to learn who you are and how you like to be communicated with. But it’s still reactive. It’s not responsive to your needs. And it certainly isn’t proactive.
From Interfaces to Personalized Interactions
In the past, and with many legacy products used every day all over the world generating billions of dollars in revenue, the message has been “Here’s our interface. Invest time and you’ll unlock value. Let me teach you how”.
Now, we have the rush to conversational AI. Just this week, the Wall Street Journal stated that “nearly every software-as-a-service company is now selling agent tools to their customers”. But these are bolt-on solutions created after the fact. They are reactive to the user’s needs, as they are expressed explicitly. But as the MIT finding recently cited by Fortune that 95% of enterprise AI solutions will fail by the end of the year indicates, it is because of the continued learning gap. Tools are too hard to learn for individuals and organizations. Crucially, generic tools like ChatGPT “stall in the enterprise because they don’t learn from or adapt to workflows”. Essentially, the model is “Tell me what you want, and I’ll give it to you instantly. But you have to tell me. And all you have is this chatbox.” And that’s not going to cut it. The learning curve isn’t just steep, it’s unacceptable.
Going forward, we need to move to a model of “I know you….here, let me help.” We need to combine with design-led principles that have served us well in the past the power of inference to proactively adapt the user experience to enable each user accomplish their goal in a personalized way.
For PLG leaders, this means adoption loops look entirely different:
Interfaces shrink. Prompt fields and conversational UIs replace complex menus.
Education disappears. The best products reveal functionality through user intent, not tours. And each user gets a unique experience personalized to their needs.
Retention shifts. Stickiness isn’t about mastering features, it’s about trusting the product to deliver without instruction. And each user journey can be different.
The products that scale in the AI-native world will be the ones that act like partners, not platforms. This is a tall order and will require deep exploration and experimentation. The first iteration of which we’re seeing with the advent of Agents. But we have a long way to go.
Onboarding as Discovery vs. Education
Onboarding used to be a core PLG lever. It was where you taught users how to get started, introduced features, and nudged them toward activation.
At Slack, we invested in explaining integrations, showing teams how to create them in channels and use email to bring colleagues into the channel. That worked and Slack grew explosively, and in many enterprise accounts, we out-competed Microsoft Teams.
But in 2025, those same tours would be attention killers.
At Outlook mobile, we designed beautiful guided tours. We soon ripped them out as they were actually driving down activation. It worked and we started to see the climb up towards our first 60M MAU milestone. We know they don’t work.
Why? Because onboarding as education assumes patience. And patience is gone. Userpilot still advocates in-app educationals, but warns that they need to be broken down to bite-sized chunks. But that’s missing the point and not looking forward to what users really expect.
AI Flips the Script - Product discovers the user’s intent
We need to harness the power of AI to deliver the experience and value AI has taught users to expect. We need to catch up with our users. This means that we need focus more than ever on the user’s intent. And use the product and onboarding moment to discover it.
This is the single biggest shift for PLG leaders. Onboarding is no longer about showing what your product can do. It’s about figuring out what the user wants to do, instantly.
This shift is made possible by the range of signals every product can now observe. Explicit inputs like search queries or button clicks, implicit behaviors like navigation paths or dwell time, and contextual information like user role, account stage, or connected data all become raw material. By combining these signals with lightweight rules, semantic matching, and AI models, products can predict what a user wants with increasing accuracy—whether it’s creating a card, importing a dataset, or setting up a recurring workflow.
The interaction model also changes. Instead of long-form tours or 20-minute training videos, intent discovery should feel conversational and lightweight: a one-line prompt, a smart default, or a pre-filled option that can be corrected with a single click. When confidence is high, the product can even “ghost” an outcome. Like pre-composing a chart or issuing a test card, and simply ask the user to confirm. Trust is built not through lengthy education, but through fast, accurate suggestions that prove the system understands you.
Done well, this turns onboarding from education into discovery. The product doesn’t ask users to learn; it shows it has already learned about them. Time-to-value shrinks to seconds, the learning curve disappears, and adoption loops accelerate because users feel like the system is anticipating needs rather than demanding attention. That’s the AI-native standard, and the new measure of whether a PLG motion is competitive.
Onboarding is no longer about teaching users what your product can do. It’s about proving your product already understands what they want to do.
Competition is Invisible and Instant
In the old PLG playbook, your competition was another SaaS tool and the battle lines were clear. When we were fighting to unseat Apple Mail, the customer pulled out a spreadsheet of all the features we “have to build”. And then at Slack in head-to-head duels at enterprise customers against Microsoft Teams, Microsoft used to love turning the conversation to a simple spreadsheet of check-boxes. Showing all of the features they’d copied (but poorly implemented).
As I used to say at Outlook mobile and at Slack, “be competitor aware, not competitor obsessed”. And particularly in the case of Teams, understand the whole picture, not just what’s in front of the user. (But that’s another article.)
If you’re worried about feature gaps now, you’re in even worse shape in a world where AI-native apps don’t care about feature matrices. Today, your biggest competitor is ChatGPT, Claude, or Perplexity solving the user’s problem in 30 seconds. Without signup, without a credit card, without onboarding.
Google is the first obvious victim. As Google execs show concern that users don’t want to click, read, click, click, click to find an answer. They just want the answer. But these general purpose AIs are also coming for you. The flexibility and power hidden behind the prompt box is real.
This is the existential threat for PLG leaders. The user’s question has shifted:
Old: “Which SaaS product should I use to solve this problem?”
New: “Why can’t I just describe what I want and have it work?”
Implications for PLG
Differentiation moves up the stack. Competing on features has always been a race to to the bottom. AI makes this instant. You must compete on data, workflow integration, and outcomes. Focus on the user intent, the outcome, and the value. And deliver it beautifully.
Moats look different. It’s no longer features vs. competitor X. It’s whether your product is the fastest, most trustworthy orchestrator of intent into outcome.
Distribution must change. SEO, ads, and freemium are still useful, but the real battle is being the invisible layer inside workflows (APIs, agents, embedded AI copilots).
The competitive battlefield is no longer a Gartner quadrant. It’s the user’s brain in the split second they decide whether to open your app, or just ask ChatGPT.
Conclusion: PLG is Dead. Long Live PLG.
The old PLG playbook isn’t irrelevant, it built SaaS as we know it. I’ve lived its power firsthand. At Microsoft, reducing friction was the lever that scaled Outlook mobile adoption against entrenched giants like Google and Apple. At Slack, carefully crafted integrations and thoughtful onboarding prompts unlocked massive enterprise adoption. Across both Microsoft and Slack, freemium economics delivered efficiency in growth that would have been impossible through sales-led motions alone. And in both cases, the combination of go-to-market strategy and deep product integration created durable category leadership.
But what worked then won’t work now.
In the AI-native world, the rules of growth have fundamentally shifted. Time-to-value must be instant, not measured in days or sessions. Learning curves must be inverted, with products adapting to the user instead of the other way around. Onboarding must become discovery, revealing intent rather than teaching features. And most importantly, competition is invisible and instant. The threat is no longer the SaaS rival across the street, but an AI tool that can solve the same problem in seconds without sign-up or onboarding.
This is both terrifying and exciting. For founders, product leaders, and investors, it means rethinking every assumption we’ve held about PLG.
That’s exactly what I’ll be sharing at the Product-Led Summit in San Francisco on September 16th, where I’ll present “PLG is Dead. Long Live PLG: How AI is Rewriting the Rules of Product-Led Growth.”
If you’re attending, I’d love to continue this conversation in person. And if you’re reading this later, know this: the companies that thrive in the next decade won’t just use AI. They’ll rebuild PLG itself around it.
In the AI era, PLG doesn’t guide users to value. Products must deliver it before they ask.
💡 Enjoyed this essay?
👉 Subscribe free to get new posts in your inbox each week.
🔒 Paid subscribers unlock full archive access, including past essays, deep-dive playbooks, and private frameworks I use with SaaS founders and investors.
Keep reading with a 7-day free trial
Subscribe to Beyond the Build to keep reading this post and get 7 days of free access to the full post archives.