What GitHub Copilot's $2B run taught us about how AI is rewriting the product-led growth playbook

"Your trial ends in 3 days. Would you like to upgrade?"

I stared at the GitHub Copilot notification, realizing I hadn't even thought about the subscription cost for weeks. The AI coding assistant had become so seamlessly integrated into my workflow that canceling felt like voluntarily removing my brain's autocomplete function. Six months later, when GitHub announced Copilot had reached over $2 billion in annual recurring revenue, I understood why: they had cracked the code on AI-powered product-led growth.

GitHub Copilot showed how AI fundamentally changes the product-led growth equation, turning products from tools users consciously choose into intelligent systems that become nearly impossible to live without.

Why traditional PLG is breaking down

Product-led growth dominated the last decade with a simple formula: create an intuitive product, offer a generous free tier, let users experience value quickly, then convert through usage-based limitations. Companies like Slack, Zoom, and Notion built billion-dollar businesses this way.

But AI is disrupting every assumption underlying traditional PLG strategies.

Time to value isn't a moment anymore. Traditional PLG focuses on getting users to their first "aha moment" as quickly as possible. AI products deliver value continuously and compound over time as they learn user patterns. The value isn't in a single moment; it's in accumulating intelligence that makes the product more valuable each day.

Freemium economics get complicated. When your product learns from each user interaction, giving away free usage means subsidizing AI costs while training models that primarily benefit paid customers. The unit economics that worked for storage-based freemium tiers collapse under AI computational costs.

Growth happens differently. Traditional PLG relies on word-of-mouth and viral loops. AI products often spread through productivity gains that are hard to attribute but impossible to ignore. Colleagues notice when you're coding faster or making smarter decisions.

How AI creates new growth superpowers

While AI breaks traditional PLG models, it also creates entirely new growth mechanisms that are more powerful than anything we've seen before.

At Microsoft Build 2025, the company announced over 50 AI-powered tools, including new coding agents for GitHub and multi-agent systems. Google I/O 2025 revealed that AI usage across Google's products climbed from 9.7 trillion tokens in April 2024 to over 480 trillion tokens in April 2025—a 50x increase in just one year. These numbers show the exponential adoption patterns that AI enables.

AI creates habit-forming workflows, not just features. GitHub Copilot doesn't just autocomplete code; it changes how developers think about programming. Users report feeling "cognitively incomplete" without it. This isn't feature stickiness—it's workflow dependency.

Teams adopt through collaboration, not marketing. As AI learns individual user patterns, it becomes increasingly valuable to that specific person. When teams collaborate, they experience AI capabilities through shared work products, creating organic bottom-up adoption pressure.

Value compounds with usage. Traditional products deliver consistent value per use. AI products become more valuable with each interaction, creating positive feedback loops that increase switching costs exponentially.

Three patterns driving AI product success

Looking at successful AI product launches, three clear patterns emerge:

1. The workflow integration model (GitHub Copilot) Copilot succeeds because it integrates into existing developer workflows without requiring behavior change. Developers don't need to learn new interfaces or change their coding habits. The AI simply makes their existing process faster and smarter.
2. The personalized intelligence model (NotebookLM) Google recently launched NotebookLM as a mobile app, allowing users to synthesize documents into podcast-style summaries they can interact with. The product becomes more valuable as users feed it more personal documents and context.
3. The enterprise viral model (Microsoft 365 Copilot) Microsoft announced new multi-agent capabilities where AI systems work together as teams. When one team member uses AI to create better presentations or analysis, it influences adoption decisions by colleagues and management. AI productivity gains are often visible to others, creating organic viral loops within organizations.

A personal lesson in AI product adoption

A few semesters ago, I was working on a data analytics platform when our team integrated AI-powered insights. Initially, we followed traditional PLG principles: prominent AI feature placement, onboarding flows highlighting AI capabilities, and usage metrics tracking AI engagement.

The results were disappointing. Users would try AI features once or twice but wouldn't develop sustained usage patterns.

Then we tried something different. Instead of promoting AI features, we embedded AI intelligence into existing workflows. When users performed routine analysis tasks, the system would quietly suggest additional insights, identify potential data quality issues, or recommend relevant visualizations.

Within three months, user engagement with AI-powered features increased substantially, but more importantly, overall platform stickiness improved dramatically. Users weren't consciously choosing to use AI, but they were benefiting from intelligence that made their existing work more effective.

The breakthrough: successful AI-PLG is about making existing product experiences more intelligent.

New metrics for the AI era

Traditional PLG metrics like activation rates and time-to-value need updating for AI products:

• Intelligence adoption rate: How quickly does the AI learn enough about individual users to provide personalized value? This is more predictive of retention than feature adoption rates.
• Workflow integration depth: How many steps in a user's workflow involve AI assistance? Deep integration across multiple workflow steps creates stronger retention than single-point AI features.
• Value accumulation: How much does each session contribute to future session value? AI products should show compounding returns over time.
• Organic discovery: Are users finding new AI capabilities through their existing workflows, or do they require explicit feature promotion? Organic discovery suggests successful AI integration.

What's coming next

Microsoft's Build 2025 theme of "the open agentic web" and Google's introduction of "Agent Mode" in Gemini suggest the next evolution: products that don't just assist users but actively work on their behalf.

Google's Agent Mode can handle complex tasks like apartment hunting by accessing multiple services and scheduling tours. This represents a shift toward autonomous AI that handles complete workflows rather than just individual tasks.

This agent-first approach will likely create even stickier product experiences, where users delegate entire categories of work to AI systems that become deeply integrated into both personal and professional workflows.

The companies that figure out how to combine the viral mechanics of traditional PLG with the intelligence accumulation and workflow integration of AI will build the next generation of category-defining products. GitHub Copilot was just the beginning.

The future of product-led growth is about building products that users choose to use. It's about building intelligent systems that users can't imagine working without.