Think about what happened to the corner video-rental store when Netflix arrived. It wasn’t that Blockbuster made bad tapes or hired the wrong staff — it’s that the ground beneath the entire business model shifted overnight. Customers didn’t need to drive to a shelf anymore. The problem the store solved had been re-solved more cheaply, more conveniently, and at massive scale by something that hadn’t existed five years earlier.
That is roughly what is happening right now to a generation of software startups — companies that raised money, hired teams, and built products in the years before November 2022, when OpenAI launched ChatGPT and generative AI (AI systems that can produce text, code, images, and analysis from a simple prompt) moved from research labs into everyday business software.
What Is It?
The phrase “pre-ChatGPT startup” refers to any technology company whose core value proposition — the main thing it does that customers pay for — was built on capabilities that large language models (LLMs) can now replicate or dramatically undercut. LLMs are AI systems trained on vast amounts of text that can understand and generate human-like language. They underpin tools like ChatGPT, Google Gemini, and Claude.
Before these models became widely available, a startup that automated customer-support emails, summarised legal documents, translated content, or generated first-draft marketing copy could charge a healthy subscription fee because doing those tasks well required significant human effort. After 2022, the same tasks became something any company could approximate with an off-the-shelf AI tool, sometimes for a few dollars a month. That is the “disruption” part of “disrupted or dead.”
The Real Mechanics
Before the shift: a software founder identified a narrow, painful business problem — say, extracting key clauses from supplier contracts — and built a purpose-built tool to solve it. The moat (the competitive advantage that kept competitors out) was often the proprietary workflow, the trained model, or simply being first. Customers paid because building the same thing in-house was expensive and slow.
What changed: when foundation models — think of them like a very powerful, general-purpose engine that any developer can drop into their own product — became commercially available via simple APIs (application programming interfaces, the connectors that let different software talk to each other), the cost of solving those same narrow problems collapsed. A developer at a large company could now spin up a comparable solution in an afternoon using OpenAI’s API or a rival service. The specialist startup’s multi-year head start evaporated.
Think of it like this: imagine you spent three years perfecting a handmade bolt-tightening jig for a specific car part. You’ve refined the grip, the angle, the torque. Then a general-purpose robotic arm arrives on the factory floor that can do the same job — and a hundred others — after a 20-minute software configuration. Your jig isn’t broken. It’s just been leapfrogged.
Why now specifically? Three forces converged almost simultaneously. First, the underlying models became dramatically more capable with GPT-4 and its peers in 2023. Second, the cost of running inference (actually using an AI model to process a query) dropped sharply as competition between model providers — OpenAI, Anthropic, Google, Meta, Mistral — intensified. Third, enterprise buyers, once cautious about AI, began allocating real budget to AI-native tools, creating pressure on incumbents to integrate or lose contracts. As we’ve explored in our analysis of Nvidia’s push into the $200 billion agentic AI market, the infrastructure layer is now mature enough to support widespread enterprise deployment — which accelerates exactly this kind of disruption.
There is a compounding dynamic here that is easy to miss: the same wave of AI investment that is killing pre-ChatGPT startups is simultaneously funding their replacements. Venture capital that once flowed to narrow-task automation tools is now concentrating in AI-native “vertical agents” — software that doesn’t just assist with a task but autonomously completes it end-to-end. The net effect is that the startup ecosystem is not shrinking; it is turning over. The graveyard and the nursery are right next to each other, which is why aggregate VC numbers look stable even as individual company mortality rises.
Why Does It Matter?
For executives watching from the sidelines, this story has two strategic implications that go beyond startup gossip.
First, vendor risk is now an AI-readiness question. If your company relies on a SaaS (software-as-a-service) tool built on a narrow AI capability that existed before 2022, it is worth asking whether that vendor has a credible AI-native roadmap — or whether they are running on borrowed time. A vendor that fails quietly mid-contract creates data migration headaches and operational gaps. Data quality and continuity become acute concerns when a platform disappears mid-integration.
Second, the bar for what counts as a defensible product has risen. A startup that can be replicated by a weekend prompt engineer using commodity AI tools is no longer a startup — it is a feature waiting to be absorbed by a larger platform. The companies surviving this shakeout share a common characteristic: they have proprietary data, deep workflow integrations, or network effects that a general-purpose model cannot simply absorb. This mirrors the broader point that agentic AI systems are increasingly capable of replacing multi-step workflows, not just single tasks.
Edge Cases
Not every pre-ChatGPT startup is doomed, and it is worth being precise about where the disruption is sharpest and where it is softer.
The disruption hits hardest in categories where the product’s core value was language processing at scale: content generation, basic chatbots, simple summarisation, template-based document creation. These were never truly defensible; they were arbitrage plays on the gap between human labour costs and AI capability — and that gap has closed.
It hits softer in categories where the value is data, relationships, or regulatory complexity. A compliance software company that has spent a decade curating financial-regulation databases and building integrations with core banking systems is not easily disrupted by a general-purpose LLM. The data is the moat, not the language processing on top of it.
There is also a hardware frontier worth watching. Breakthroughs like light-powered valleytronics chips could eventually make AI inference so cheap and fast that even more product categories face commoditisation — but that transition is years away from commercial scale.
Common Misconceptions
Misconception 1: “This is just normal startup churn.” It isn’t, quite. Normal startup failure rates reflect bad execution, poor product-market fit, or running out of cash. What is unusual here is that many of the companies being disrupted had product-market fit and paying customers. The problem is not that they failed — it is that their category was invalidated beneath them. That is a different phenomenon with different lessons.
Misconception 2: “AI-native startups are automatically safe.” Building on top of a foundation model creates its own set of risks. If your entire product is a thin wrapper around GPT-4 — meaning your app mostly just passes user inputs to OpenAI and returns the response — you are just as exposed as the pre-ChatGPT startups were. OpenAI, Google, or Anthropic can add your feature to their own product at any time. True defensibility requires something the foundation model provider cannot easily replicate: proprietary training data, deep vertical integrations, or a network that grows more valuable with each user.
Misconception 3: “Big companies are immune.” Enterprise size slows the process but does not stop it. Large organisations that have embedded pre-ChatGPT tools deeply into their operations face their own version of the same problem: technical debt (the accumulated cost of outdated or inefficient technology choices) and the organisational inertia that comes with ripping out an integrated system. The disruption is coming for them too — just on a longer timeline. The lag between AI capability and organisational adaptation is a well-documented pattern across industries.
What the Pre-ChatGPT Startup Story Is Missing
The “disrupted or dead” narrative is real, but it leaves some important threads hanging.
The acqui-hire angle. Many pre-ChatGPT startups are not dying so much as being absorbed — their teams and codebases acquired by larger AI-native companies or cloud platforms at fire-sale prices. This is less visible than a shutdown but represents a significant transfer of talent and intellectual property that reshapes the competitive landscape. The story rarely tracks where the teams and technology end up.
Geographic and market variation. Startup disruption by AI is not uniform across geographies. Markets with less developed AI tooling ecosystems — parts of Southeast Asia, Latin America, and Africa — may still have a viable window for startups solving problems that US and European competitors have already written off as commoditised. The “disrupted or dead” headline is largely an American and European story.
The regulatory wildcard. AI regulation in the EU (under the EU AI Act), and emerging rules in the US and UK, could re-raise the compliance cost of deploying foundation models in high-stakes sectors — potentially giving a reprieve to specialist startups that have already navigated those regulatory hurdles. This dimension is almost entirely absent from the current disruption narrative.
Where to Learn More
If this topic has caught your attention and you want to go deeper, here are some substantive next steps:
- Foundation model capabilities: The OpenAI platform documentation gives a clear picture of what developers can build today with commodity AI APIs — useful for benchmarking which of your vendors might be vulnerable.
- EU AI Act: The European Commission’s AI policy hub tracks the evolving regulatory framework that will shape how AI tools can be deployed in enterprise contexts.
- Competitive dynamics: Our deep-dive on AI displacing cybersecurity professionals explores a parallel disruption story in a high-stakes vertical — useful context for understanding how the pattern plays out sector by sector.
Three Things to Track
- Foundation model pricing announcements. Every time OpenAI, Anthropic, or Google cuts inference costs, the radius of “AI can do this cheaply” expands. Watch the API pricing pages of the top three model providers for downward moves — each one effectively invalidates another tier of pre-ChatGPT products.
- Enterprise SaaS renewal rates in AI-adjacent categories. Publicly traded SaaS companies in document automation, customer-support tooling, and content generation will report Q3 and Q4 renewal data in coming quarters. Declining net revenue retention in these categories will be the clearest signal of how fast enterprise buyers are replacing legacy tools.
- Regulatory rulings under the EU AI Act. The first enforcement actions or compliance guidance targeting foundation-model APIs in high-risk sectors (healthcare, finance, HR) could meaningfully shift the build-vs-buy calculus for enterprises — and give specialist compliant-by-design startups an unexpected competitive window.
