“Hallucination,” “alignment,” “confidently wrong,” and, well, “crazy” are some of the cautionary words that we use to describe a fundamental problem with large language models (LLMs). These models are “reasoning machines” and NOT “fact machines.” So to build useful and differentiated applications, developers need to incorporate LLMs into broader architectures that include access to data sources and APIs. Open Source AI and the proliferation of models it is enabling also adds to the complexity of that stack. Powerful open-source tooling and an array of plugins have emerged to make all this possible. But today, few development teams have the skill to assemble those pieces into a working demo, let alone a scalable, secure, and efficient production application.
If you believe (we do) that the best ideas can come from anyone, the complexity and friction of building with LLMs today is a major problem. Because higher complexity and friction mean fewer people can bring their ideas to life. That’s why we are so excited to announce our investment in Fixie, a platform as a service that makes it easy for developers to build LLM-based apps that work with the broader architectures those apps need to be useful and differentiated.
Fixie was founded by Matt Welsh, a veteran of two Madrona-backed startups (Xnor and OctoML) and Google, and he is a former Harvard CS professor. Matt has argued that LLMs and generative AI herald “the end of programming” and the beginning of something new. With Fixie, Matt has assembled a team with experience from Google, Apple, Shopify, and beyond to give us all a glimpse of that something new.
Fixie makes it easy for developers to assemble complex LLM application stacks into useful, differentiated applications. It does that with “agents” — powerful units of compute that combine the reasoning abilities of LLMs with memory, access to data sources, and APIs. Agents are reusable, composable, and easy to create because they use LLMs under the hood. You can create Fixie agents with your programming language of choice, a LangChain script, or even with natural language. Then, you can train the agents to interact with data sources and APIs with just a few examples.
Fixie-enabled apps can apply the reasoning of LLMs to data retrieved from public and proprietary enterprise databases. Likewise, for taking LLM-informed action via private and public APIs. Fixie agents can call your favorite ChatGPT plugin, and you can also use Fixie to build new ChatGPT plugins to empower other applications. Fixie is model-agnostic (works with GPT-4 out of the box) but supports any model or ensemble of multiple models a developer wants to use. All of this is managed by Fixie in the cloud to seamlessly move from hackathon to secure, scalable, and efficient production.
Madrona has backed AI founders working to make the future happen faster for over a decade. We backed the Allen Institute for AI (AI2) in 2012 to tackle some of the world’s toughest challenges with AI, producing a deep bench of AI talent and companies. Over the years, we have backed companies like OctoML, Turi (acquired by Apple), Lattice (acquired by Apple), Algorithmia (acquired by Datarobot), and Xnor (acquired by Apple). Some of our recent investments in this area include Runway and Numbers Station. (Matt Welsh was an early employee at Xnor.ai and OctoML, so he has the deep engineering chops we value.)
We saw in all those founders what we see in the Fixie team: impatient optimism, deep customer empathy, and a rare density of talent. We could not be more excited to work with Matt Welsh this 3rd time and to welcome Zach Koch, Justin Uberti, Hessam Bagherinezhad, and the rest of the Fixie team to the Madrona family. We’re looking forward to the journey to bring the power of LLMs and generative AI into more developers’ hands, thereby bringing those developers’ great ideas to life.
Today, Fixie is announcing the launch of its developer preview along with its seed funding round. You can learn more about Fixie by reading the Fixie team’s blog post here. You can sign up for early access to the Fixie platform here.