73% of AI Startups Found Repackaging Existing Models, Sparking Industry Debate

A recent Medium article, titled "I Reverse-Engineered 200 AI Startups. 146 Are Selling You Repackaged ChatGPT and Claude with New UI," has ignited significant discussion within the AI and tech communities after reaching the front page of Hacker News. The investigation, conducted by Teja Kusireddy, a Founding Engineer at GetOnStack, revealed that a substantial majority—73%—of the 200 funded AI startups analyzed are primarily utilizing third-party large language model APIs, such as those from OpenAI and Anthropic's Claude, with minimal proprietary innovation.

Kusireddy's findings indicate that many of these companies are essentially "wrapper companies," adding a new user interface or minor additional steps to existing powerful models. The author also highlighted a concerning instance where at least one startup was imposing a 1000x markup on the underlying model API costs, suggesting a significant disparity between perceived value and actual technological contribution. "For Investors: You’re funding prompt engineering, not AI research. Adjust your valuations accordingly," Kusireddy stated in the article, directly addressing the investment community.

The article's methodology involved monitoring network traffic, decompiling code, and tracing API calls across a sample of funded AI startups. This forensic approach uncovered patterns such as companies claiming "proprietary models" that were, in reality, GPT-4 with added layers, or "revolutionary natural language understanding engines" that simply used system prompts to guide existing APIs. Kusireddy emphasized that while wrappers can accelerate time-to-market and demonstrate clear user value, they often lack durable competitive advantages.

Industry reactions have varied, with some acknowledging the validity of Kusireddy's observations and others pointing out the practicalities of building on foundational models. The article has prompted a broader conversation about genuine innovation versus repackaging in the rapidly expanding AI sector. Experts suggest that true long-term success for AI companies will depend on building "moats beyond the wrapper," focusing on workflow integration, proprietary data, distribution channels, and real-world actions rather than just access to large language models.

The revelations underscore a critical juncture for the AI startup ecosystem, urging investors, customers, and developers to scrutinize the underlying technology and business models more closely. As Kusireddy advised, "If your core advantage is 'we call GPT-4 well,' you do not have an advantage." The discussion continues to evolve around what constitutes valuable AI development and how to foster sustainable growth in a market increasingly reliant on powerful, externally provided AI infrastructure.