By: Thomas Stahura
It was my second desperate pivot, and it made so much sense at the time. An AI marketplace I thought! A site where users can submit and monetize their prompts and use cases.
Turns out, a chat interface is much more intuitive than searching a giant list of prompts.
So last year, when I heard OpenAI killed 100,000 startups with the launch of its GPT store, I was justifiably skeptical. But it got me wondering: How many companies has OpenAI actually killed? And more broadly, how has AI affected the tech landscape 2 years into the fourth industrial revolution?
Let’s start with the most visible disruption.
Devtool and edtech companies that once seemed untouchable are crashing back down to earth. Since 2022, Stack Overflow, a question-and-answer platform for developers, lost about 5 to 15% of its web traffic each year. In response, it launched OverflowAI in the summer of 2023. Despite the push, Stack Overflow’s decline has not slowed down. Chegg, a study and homework help platform, rolled out CheggMate in spring 2023. Since then, its stock plunged 97%. Coursera, another edtech company, launched its AI-powered Coursera Coach last year. The stock is down 85% since 2021.
Meanwhile, AI is creeping into the design world: Adobe launched Firefly, its AI image generator; Canva rolled out Canva Code, its text-to-design tool; and Figma followed with Figma Code, its own version of text-to-design. Unlike education or developer tools, the design sector is still growing, but that likely won’t last for long. Large language models can now generate full applications from a simple prompt.
Lovable, on its home page, advertises itself as a Figma competitor. For those still designing by hand, it added an "import from Figma" button. The once-dominant design firm — which nearly sold for $20 billion in 2023 — is now reduced to a button on a rival's site. Figma responded by launching its own AI dev tool, Figma Code, and issued cease-and-desist letters to Lovable and others over their use of "Dev Mode," a term Figma trademarked in 2023.
It’s getting ugly for the companies not named OpenAI.
Speaking of, OpenAI’s image generator now produces nearly perfect text and designs. Using 4o feels like how Photoshop should work — and Adobe better be taking notes.
AI labs are racing toward models that can handle every modality, and businesses are restructuring their products around them. When every product works like a text-to-anything tool, how will users tell them apart?
Honestly, besides UI and mindshare, what are the differences between Lovable, Bolt, Chef, Github Spark, v0, Firebase Studio, AWS App Studio, Cursor, Windsurf, Claude Code, Codex, Figma Code, or Canva Code? (And that's just the tip of the iceberg.) Some may use different models, but even that layer is close to being commoditized.
So how are entrepreneurs supposed to stand out?
The new frontier in the digital world will probably be vertical AI, or what we call SaaS 3.0. These are tools built for specific industries, workflows, companies, or even individual users. Here, differentiation does not come from the model or UI, but from data, domain expertise, and deep trust.
Rohan D’Souza, founder of Avante, a health benefits admin platform and Ascend portfolio company recently wrote in a post:“The model is the tiniest piece of a much larger enterprise stack required to actually deliver value.”
In other words, the real moat is not the model itself. It’s the safety, reliability, domain-specific workflows, and trust built around it.
I believe the digital frontier is only half the story. For decades, the most dramatic technological shifts happened on screens and servers. As Marc Andreessen famously put it: "Software is eating the world." It took a while, but AI is breaking out of code and moving into the physical world — biotech, robotics, manufacturing, logistics, and more.
AI in the physical world is far more defensible. The machines it runs are harder to replicate, and the technical nuances go deeper than traditional software alone. (Ascend labels this category Frontier AI).
Despite OpenAI's partnership with Anduril, the demand for homegrown physical tech alternatives is only growing. For instance, in 2022 the American Security Drone Act banned federal agencies from using Chinese-made drones and parts. Around that time, some of my college friends were running Uniform Sierra, an aerospace startup focused on building high-quality drones in the U.S. They scaled with a 3D printer farm as demand surged, and the company was recently acquired. More startups, like Seattle-based drone startup Brinc, are reshoring their manufacturing apparatus.
So did OpenAI kill 100,000 startups? Probably a few thousand. Mine for sure. But in my defense, I built a chat app, a marketplace, and a social media site before OpenAI did. I have the right ideas. I could have kept going — and I still probably would have been steamrolled.
My chat app worked because there were no others like it at the time. I knew back then it wouldn't last. LLMs were too good to stay secret, and OpenAI would productize them better than I could using its API. I knew I had to differentiate. Now chat apps are a dime a dozen.
Differentiation mattered then. It matters even more now, especially with trillion-dollar tech giants pivoting their entire product suites into AI. Timing might get you started, but differentiation keeps you going.