Investment Thesis

We bet on the stack before the stack had a name.

Since 2020, Hearthstone has backed founders building the technical substrate for AI-native software: the pipelines, memory layers, orchestration frameworks, execution engines, and developer surfaces that make AI applications possible at scale.

The Agentic Moment

The first wave of applied AI adoption — 2022 to 2024 — was dominated by the race to integrate foundation models into existing products. What emerged from that wave is something more interesting: a nascent infrastructure category for AI agents. LLM pipelines. Memory stores. Orchestration frameworks. Action execution layers. Agent routing systems. Usage-based billing infrastructure for AI workloads.

This is the layer we focus on. Not the foundation models (that battle is being fought at a scale we don't play at), and not the consumer applications (too wide, too subject to platform risk). The infrastructure layer between the two: the tools developers reach for when they are building AI-native products for the first time.

We believe this layer will produce 10 to 20 significant companies over the next five years. We intend to be shareholders in several of them from the beginning.

What We Look For

We are not generalist investors who happen to be interested in AI. Both Hearthstone partners spent years as engineers and product leads building AI-adjacent systems before we wrote a single check. We know what the inside of this problem looks like — the tooling gaps, the integration headaches, the places where a good abstraction would save 200 engineering hours.

That shapes how we evaluate companies. We are drawn to founders who have a precise diagnosis of a real technical gap — not just 'AI is the future' but 'here is the specific thing that does not exist yet and here is why we are the team to build it.' We want the thesis to be tight enough to be wrong.

Infrastructure depth

Products that sit below the application layer and enable the layer above. Primitives over features.

Developer-led adoption

Products that developers want to use before they are told to. Bottom-up distribution signals genuine utility.

Technical founding teams

At least one founder who has shipped production AI systems before. Domain knowledge that is hard to acquire.

Pre-category clarity

We are most useful at Seed, before there is established pricing or a clear category name. After Series A, we are probably late.

The Funds

Hearthstone Fund I

$38M Closed June 2021

First fund; established the seed-stage AI infrastructure thesis. Eight portfolio companies to date.

Fully deployed

Hearthstone Fund II

$47M Closed February 2024

Second fund; continuing the seed thesis with expanded coverage of agentic systems and applied AI platforms.

Currently deploying