Why revisit disruption now?
In Level 1 you reviewed Christensen's model. Incumbents improve their products faster than most customers need, so they overshoot the market. A challenger then enters with something cheaper or simpler that is "good enough". This happens either at the low end of an existing market or in a new market made up of people who were not buying at all. The incumbent ignores the challenger because its margins look unattractive, and then the challenger moves upmarket.
AI does not replace this model. It changes the shape of the curves inside it. This lesson covers three changes, two case studies and a first test you can run on your own market.
Shift 1: The cost to serve falls sharply
In Christensen's model, the challenger usually wins on cost. AI pushes that further because it lowers the cost of cognitive work: reading, writing, answering, classifying and drafting.
- Evidence: a16z's November 2024 "LLMflation" analysis estimated that, at a fixed capability level, the price of LLM inference was falling about 10x per year. For a performance level GPT-3 reached in November 2021 at roughly $60 per million tokens, a comparable model cost roughly $0.06 by late 2024. Treat the exact rate as an estimate. The direction is clear.
- What it means: work that once needed a salaried person or a costly expert can now cost cents per task. A challenger can profitably serve customers the leader cannot afford to serve.
DO: In your market, list the three most expensive human tasks in delivering the product, then estimate what share of each is reading, writing or answering.
Shift 2: "Good enough" arrives faster
In the classic model, a disruptor's performance curve climbs over many years. With AI, much of the improvement comes from the underlying models, and they improve every few months. Two consequences follow:
- The gap closes faster. A product that is too weak for mainstream customers today may be good enough within one or two model generations.
- The improvement is available to everyone. The same model upgrades are on sale to the incumbent. A faster curve helps the challenger only if the incumbent's business model, not its technology, stops it from using the upgrade.
DO: For your market, ask which tasks a current model gets 80% right today. Assume those reach "good enough" soon, and plan around that.
Shift 3: Software can do work it could not do before
Earlier software automated structured tasks: forms, databases, rules. AI can handle unstructured tasks such as free text, conversation, documents and judgment calls. This creates new-market disruption: it can reach non-consumers who could never afford a human expert, or whose needs were too small or messy for traditional software.
DO: List who in your market goes without help today because it is too expensive, too slow or too complex. Those people are your possible new market.
Case studies: threat or tool?
Tool for the leader: GitHub Copilot. GitHub released Copilot as a technical preview on 29 June 2021 and made it generally available to individual developers on 21 June 2022, at $10 per month. AI made GitHub's existing product better for its existing customers, through its existing sales channel. That is sustaining innovation. The leader used AI and became stronger.
Threat to the leader: Chegg. Chegg sold homework help by subscription. On 2 May 2023 it said ChatGPT was hurting new customer growth, and its shares fell about 46–48% that day. By Q1 2025, revenue was down 30% year on year to $121 million and subscribers were down 31% to 3.2 million. Chegg cut 22% of its staff. General-purpose AI gave students "good enough" answers at little or no cost. That is low-end disruption driven by Shift 1.
A caution on "good enough": Klarna. On 29 February 2024, Klarna said its OpenAI-powered assistant was handling two-thirds of customer service chats, equal to the work of 700 full-time agents. In May 2025, its CEO told Bloomberg that the cost focus had produced lower quality, and Klarna began hiring human agents again. Lower cost to serve is real, but it disrupts only where the quality is good enough for the customer.
Run your first test: is AI a threat or a tool?
Pick one market and its current leader, then follow these steps:
- Write down how the leader makes money. For example: per seat, per subscription, per billable hour or per transaction.
- Define the job. In one sentence, say what customers pay the leader to get done.
- Apply the three shifts. Mark each one high, medium or low for this market: cost to serve, speed to good enough, new work possible.
- Ask the four sorting questions:
- Does the AI product serve the leader's best customers better, on the measures they already value? (If yes, it leans sustaining.)
- Does it start with overserved customers or non-consumers? (If yes, it leans disruptive.)
- Can the leader add it without lowering its margins or cannibalising its revenue? (If yes, it leans sustaining.)
- Would adopting it force the leader to change how it charges? (If yes, it leans disruptive.)
- Give a verdict: threat, tool or unclear. Write one sentence explaining why.
Keep your notes. Later levels build on this test.
Key takeaways
- AI changes three things in Christensen's model: a lower cost to serve, faster improvement toward "good enough" and new kinds of work that software can do.
- Faster model improvement is available to incumbents too. The challenger's edge comes from business-model conflict, not technology alone.
- AI that improves the leader's product for its existing customers is sustaining (GitHub Copilot). AI that gives overserved customers a cheaper, good-enough option is disruptive (ChatGPT vs Chegg).
- "Good enough" depends on quality. A cheaper service fails if customers will not accept it (Klarna).
- Always ask whether AI is a threat to the leader or a tool for it before you build or invest.