The generative AI frenzy created tons of startups. However, as things calm down, two business models – LLM wrappers and AI aggregators – aren't looking so hot anymore. According to Darren Mowry, a Google VP, startups using these models might be in trouble.

LLM wrappers are basically startups that take existing AI models (like Claude, GPT, or Gemini) and add a layer of product or user experience to solve a specific problem. Think of it as a startup using AI to help students study.

Mowry mentioned that if you're just relying on the AI model to do all the work, the industry isn't going to be patient with that for long. Wrapping a thin layer of intellectual property around Gemini or GPT-5 doesn't make you stand out.

Startups need serious, wide advantages that set them apart, either across the board or within a specific market, to really grow. Some examples of successful LLM wrappers include Cursor, an AI-powered coding assistant, and Harvey AI, an AI legal assistant.

It's not enough anymore to just slap a user interface on top of GPT and expect your product to take off. The real challenge is building lasting value.

AI aggregators are basically wrappers that combine multiple LLMs into one interface or API. They route queries across different models and give users access to several options. Think of Perplexity, the AI search startup, or OpenRouter, which provides access to multiple AI models through a single API.

While some of these platforms have gained popularity, Mowry's message to new startups is clear: avoid the aggregator business. He says users want built-in intellectual property that ensures they're routed to the right model based on their needs, not just because of behind-the-scenes limitations.

Having worked at AWS and Microsoft before Google Cloud, Mowry has seen this before. He compares the current situation to the early days of cloud computing when many startups popped up to resell AWS infrastructure. They marketed themselves as easier entry points with tools, billing, and support. However, when Amazon created its own enterprise tools, most of those startups got pushed out. The only ones that survived were those that added real services like security, migration, or DevOps consulting.

AI aggregators face similar challenges as model providers expand into enterprise features, potentially cutting out the middlemen.

On a brighter note, Mowry is excited about code platforms, which had a great year. Startups like Replit, Lovable, and Cursor are attracting lots of investment and customers. He also sees potential in direct-to-consumer tech, putting powerful AI tools directly into the hands of users. For example, film students could use Google's AI video generator, Veo, to bring their stories to life.

Beyond AI, Mowry thinks biotech and climate tech are also doing well. Both industries are seeing lots of investment and have access to huge amounts of data, allowing startups to create real value in new ways.