We closed Hearthstone Fund II in January 2024. The fund is $48 million and focused on the same thesis that guided Fund I: AI Agent Systems and Applied AI Platforms. We are gratified by the support of our LP base, many of whom re-upped from Fund I, and excited to continue building alongside the companies in the AI infrastructure ecosystem.
Rather than a standard press release, I want to use this post to share what we learned from deploying Fund I and how that has shaped how we are approaching Fund II.
What Fund I Taught Us
Fund I was deployed primarily between 2022 and 2023, a period that bridged the pre-ChatGPT and post-ChatGPT eras. Our earliest investments — Langbase, Superagent, Mem.ai — were made before the public conversation about LLMs had meaningfully begun. We were writing checks based on early signals: the performance of GPT-3 on specific benchmark tasks, the architectural insights coming out of academic ML research, and our conviction that the infrastructure gap between model capability and real-world deployment was going to be a durable investment opportunity.
What we got right: the infrastructure thesis. Every company we backed in the infrastructure layer — retrieval, observability, developer tooling — has seen significantly increased demand since ChatGPT demonstrated the market. What we underestimated: how fast the category would get crowded. By mid-2023, there were dozens of companies doing variations on every infrastructure theme we had identified. The companies that separated from the pack did so based on team quality, technical depth, and the quality of their early customer relationships — not on product differentiation alone.
How Fund II Is Different
Fund II is more focused on the agentic layer specifically. Our Fund I thesis included both infrastructure for LLM applications broadly and infrastructure for AI agents specifically; Fund II is concentrated on the latter. We believe the agent infrastructure category is where the most interesting and durable infrastructure companies will be built over the next five years, and we want to be the most knowledgeable and well-networked investors in that specific space.
We are also more patient. We will write fewer checks at higher conviction levels, with more time spent before the investment evaluating the team and the architecture. The pace of Fund I produced a portfolio we are proud of, but also a few investments where we moved too quickly. Fund II will have ten to twelve investments at the seed stage, compared to fourteen in Fund I.