The generative AI surge created a wave of startups, but two once-popular models are losing favor: LLM wrappers and AI aggregators.
Darren Mowry, who leads global startups across Google Cloud, DeepMind, and Alphabet at Google, warns that startups built primarily as thin layers on top of existing models have their “check engine light” on.
LLM wrappers package models like GPT, Claude, or Gemini with a user interface to solve specific problems. But simply white-labeling a model with minimal proprietary technology is no longer enough. Startups now need strong differentiation—either deep horizontal capabilities or vertical-market expertise. Companies like Cursor (coding) and Harvey AI (legal) stand out because they’ve built defensible value beyond basic model access.
AI aggregators—a subset of wrappers—combine multiple models into one interface or API, routing queries between them. Examples include Perplexity AI and OpenRouter. While some have gained traction, Mowry advises new founders to avoid the space. As model providers expand their own enterprise tools, aggregators face margin pressure and risk being squeezed out.
He compares the trend to the early days of cloud computing, when startups resold AWS infrastructure until providers built native enterprise features and cut out the middlemen. Only companies offering real added services survived.
Mowry remains optimistic about developer platforms and “vibe coding,” citing strong growth for startups like Replit, Lovable, and Cursor. He also sees opportunity in direct-to-consumer AI tools—such as Google’s AI video generator Veo—as well as in biotech and climate tech, where abundant data is unlocking new value creation.
Author: Mohammed Najem