The Five Stages of Disruption in the AI Era

Introduction: When Disruption Gets Disrupted

In January 2014, Steven Sinofsky published "The Four Stages of Disruption" at Andreessen Horowitz, outlining how disruptive technologies move through predictable phases: Disruption of Incumbent, Rapid Linear Evolution, Appealing Convergence, and Complete Reimagination. The framework drew on Clayton Christensen's work and decades of technology history to help entrepreneurs and incumbents understand where they stood in the innovation cycle.

That framework was prescient and remains useful. But something fundamental has changed. The arrival of large language models and generative AI hasn't just created another wave of disruption to analyse, it has disrupted the very dynamics of disruption itself.

The patterns Sinofsky described assumed certain constants: disruption took years to unfold, new technologies started as "toys," evolution was roughly linear, and companies had time to recognise and respond to threats. AI has compressed, accelerated, and in some cases inverted these assumptions.

This updated framework preserves what still holds true from the original while accounting for what has changed. The result is five stages, not four, because AI has introduced an entirely new phase that didn't exist before.

What Still Holds True

Before examining what's changed, it's worth acknowledging the durable insights from the original framework.

Incumbent denial remains real. When ChatGPT launched in November 2022, many enterprise software leaders initially dismissed it as a "party trick" or "stochastic parrot." Google's internal response reportedly included significant concern that management wasn't taking the threat seriously. The psychological dynamics of denial, minimisation, and belated panic haven't changed, they've just been compressed into months instead of years.

The innovator's dilemma persists. Established companies still struggle to cannibalise their own products. Microsoft's bold bet on OpenAI was the exception, not the rule, and Microsoft had less to lose than Google, whose search advertising revenue was directly threatened by conversational AI. Most incumbents still face the classic dilemma: the economics of defending existing businesses conflict with the economics of embracing new ones.

Disruption still creates new categories. Just as smartphones didn't just replace phones but created entirely new markets (apps, mobile commerce, social media), AI is creating categories that didn't exist: prompt engineering, AI safety, synthetic media production, and the emerging field of AI-human collaboration.

The Five Stages of AI-Era Disruption

The original four stages have been modified and a fifth has been added to account for AI's unique characteristics.

Stage One: Capability Shock (Replaces "Disruption of Incumbent")

The original framework assumed new technologies arrived looking like inferior toys. Digital cameras initially produced lower-quality images than film. Early smartphones couldn't match BlackBerry's email capabilities. PC servers lacked the reliability of mainframes. This gave incumbents time, often years, to observe, dismiss, and eventually respond.

AI arrived differently. ChatGPT didn't look like a toy. It could write essays, explain complex topics, generate code, and engage in sophisticated conversation from day one. Within two months, it had 100 million users, the fastest-growing consumer application in history. The "toy phase" lasted approximately zero days.

This creates what we might call capability shock: the disruptive technology arrives not as a limited alternative but as something that immediately demonstrates transformative potential. The incumbent's initial reaction isn't "that's cute but limited", it's "wait, how is this possible?"

Capability shock compresses the denial phase dramatically. There's no comfortable period where incumbents can point to the new technology's obvious limitations. Instead, the conversation immediately shifts to questions like "how fast will this improve?" and "what does this mean for our business?"

Key characteristic: The new technology demonstrates capabilities that seem to exceed what should be possible given current understanding, creating immediate existential questions rather than dismissible curiosity.

Stage Two: Exponential Expansion (Replaces "Rapid Linear Evolution")

Sinofsky described the second phase as "rapid linear evolution", the disruptor fills out their product by adding features that complete their initial vision. The path forward was well understood: you drew from the cutting-room floor of ideas and fixed earlier decisions made under urgency.

AI evolution is fundamentally different. It's exponential, not linear, and crucially, it's unpredictable in its direction. GPT-3 to GPT-4 wasn't just "faster and better", it gained emergent capabilities that weren't explicitly trained for. Models developed the ability to reason about novel problems, understand context across longer passages, and perform tasks their creators didn't anticipate.

This creates a planning problem that didn't exist before. In traditional disruption, you could roadmap your evolution: "In six months we'll add X, in twelve months we'll add Y." With AI, the next capability jump might come from a research breakthrough, a novel architecture, or simply scaling compute, and its effects are difficult to predict in advance.

For disruptors, this means your competitive advantage can evaporate overnight when someone else achieves a capability breakthrough. OpenAI's lead over competitors has been repeatedly challenged by new entrants and open-source models that close the gap faster than expected.

For incumbents, traditional competitive analysis becomes nearly impossible. You can't benchmark against a moving target when you don't know which direction it's moving or how fast.

Key characteristic: Capabilities expand in non-linear jumps with emergent properties that aren't predictable from the previous state, making traditional roadmapping and competitive analysis inadequate.

Stage Three: Immediate Convergence (Accelerated from "Appealing Convergence")

In the original framework, convergence happened after the disruptor had established dominance in their new category. Having conquered the greenfield, they looked back to pick up customers who hadn't yet switched, adopting features from the incumbent they'd displaced.

In the AI era, convergence happens immediately and simultaneously with disruption. ChatGPT launched with plugins. Claude gained tool use and computer control. Every AI company is racing to integrate with existing enterprise workflows, document systems, and communication platforms, not as a later-stage growth strategy, but as a core requirement for adoption.

This reflects a deeper truth about AI as a technology: unlike previous disruptions that replaced existing tools, AI often augments them. An AI coding assistant doesn't replace your IDE, it works within it. An AI writing assistant doesn't replace your word processor, it enhances it. This means the "replace vs. integrate" decision that used to come in Stage Three is now a Day One question.

The implication for disruptors: you can't afford to be a standalone product that requires users to change their workflows. You need to meet users where they already work.

The implication for incumbents: the window for "adding AI" to your existing product is much shorter than you think, because AI-native competitors are adding your capabilities to their products just as quickly.

Key characteristic: Integration with existing workflows is a launch requirement rather than a growth strategy, collapsing what used to be a distinct phase into the initial market entry.

Stage Four: Continuous Re-imagination (Transformed from "Complete Reimagination")

Sinofsky described re-imagination as a distinct phase where a category is rethought from first principles, returning to fundamental questions like "what does it mean to capture an image?" or "what does it mean to provide transportation?" This happened periodically, generationally, as new cohorts of builders learned from both legacy and disruptive approaches.

AI has made re-imagination continuous rather than periodic. Every major model release forces a reconsideration of what's possible. When GPT-4 gained vision capabilities, every product that dealt with images had to reconsider its assumptions. When AI models learned to use tools and browse the web, every search and research product faced fundamental questions about its value proposition.

More profoundly, AI is forcing re-imagination of categories that were never before considered "tech" categories. What does it mean to practice law when AI can draft contracts? What does it mean to teach when AI can tutor? What does it mean to create art when AI can generate images from text descriptions?

The original framework suggested that previous disruptors faced a choice: embrace the legacy market more, or keep pushing forward. In the AI era, this choice is constant. Every quarter brings new capabilities that might allow you to capture adjacent markets, or might render your current approach obsolete.

Key characteristic: Fundamental rethinking of category assumptions happens continuously rather than generationally, driven by the rapid pace of capability expansion in foundation models.

Stage Five: Commoditisation Crisis (New Stage)

This stage didn't exist in the original framework because previous technologies didn't commoditise at the speed AI does. When DeepSeek released models matching frontier capabilities at a fraction of the cost, it demonstrated a dynamic unique to AI: the core technology can be replicated and open-sourced faster than traditional moats can be built.

In traditional disruption, the disruptor built durable advantages through network effects, switching costs, or proprietary technology. Facebook's social graph, Amazon's logistics network, Apple's ecosystem, these moats took years to build and proved difficult to replicate.

AI moats are more fragile. Model capabilities commoditise within months. Open-source alternatives emerge that match proprietary performance. The techniques that create breakthrough performance get published in papers and replicated worldwide. Training costs decline rapidly with each generation.

This creates a crisis that affects both disruptors and incumbents. For disruptors, it means your technological advantage is ephemeral, you need to find moats in data, distribution, or applications rather than model capabilities alone. For incumbents, it means the AI capabilities you're rushing to add will soon be table stakes, not differentiators.

The commoditisation crisis forces a fundamental question: if the core AI technology commoditises, where does lasting value come from? Possible answers include proprietary data, user trust and relationships, regulatory approval, specialized fine-tuning, and integration depth. But none of these feel as durable as the moats of the previous technology era.

Key characteristic: Core technological capabilities commoditize faster than traditional competitive moats can be established, forcing value creation to shift from technology to data, distribution, and trust.

Comparing the Frameworks

The following table summarises how each stage has evolved:

Implications for Strategy

The original framework concluded with advice for incumbents and disruptors. That advice needs significant revision.

For Incumbents

You have less time than you think. The original framework gave incumbents a comfortable window to observe, evaluate, and respond. That window has shrunk from years to months. If you're still in the "evaluation" phase of AI adoption, you're already behind. The question isn't whether to integrate AI, it's how fast you can do it while maintaining what makes your product valuable.

Your moat may not be where you think it is. If your competitive advantage is based on capabilities that AI can replicate, analysis, synthesis, pattern recognition, content generation, that advantage is eroding. The durable moats are in places AI can't easily reach: trusted relationships, proprietary data, regulatory approval, physical infrastructure, and network effects.

Integration depth beats feature parity. You won't win by building an AI that matches the capabilities of AI-native competitors. You'll win by integrating AI so deeply into your existing value proposition that switching becomes costly. The goal is to make AI a feature of your product, not a replacement for it.

For Disruptors

Capability advantages are temporary. Being first with a new AI capability buys you months, not years. The technology will commoditise. Open-source alternatives will emerge. Your competitors will catch up. Build your business model assuming that whatever capability advantage you have today will be table stakes within eighteen months.

Data is the new moat. The models themselves commoditise; the training data doesn't. If you can build proprietary datasets, through user interactions, specialised domains, or exclusive partnerships, you have something that can't be easily replicated. This is why every AI company is racing to lock in data sources.

Trust is underrated. As AI capabilities become more powerful and commoditised, the differentiator becomes who users trust to deploy those capabilities responsibly. Building a reputation for safety, reliability, and aligned values is a competitive advantage that can't be open-sourced.

Conclusion: Living in Permanent Beta

The original framework assumed disruption was a process with a beginning, middle, and end. Companies could identify where they were in the cycle and plan accordingly. The AI era has turned disruption into a permanent state. There's no stable endpoint to plan toward, only continuous adaptation.

This isn't entirely new. The tech industry has always moved faster than others. But AI has accelerated the pace beyond what previous frameworks anticipated. The question is no longer "how do we respond to disruption?" but "how do we build organisations that can continuously adapt when the ground is constantly shifting?"

Sinofsky's 2014 essay concluded that "business and product development are social sciences" with no laws of nature and nothing certain to happen. That remains true. Patterns are signposts, not prophecies. The framework offered here is a tool for thinking, not a guaranteed prediction.

But one thing does seem clear: the companies that will thrive in the AI era are those that treat disruption not as an event to survive but as an environment to inhabit. They will be the ones that can reinvent themselves continuously, not just once. They will be the ones that build lasting value in the spaces that AI can't easily commoditise: trust, relationships, and the human judgment that guides how powerful technologies are deployed.

The relentless march of technology change comes faster than you think. In the AI era, it comes faster than anyone thinks.

This essay updates and extends Steven Sinofsky's "The Four Stages of Disruption" (2014). The original framework remains essential reading for understanding the dynamics of technological change.