$100M Anthology Fund 60-Day Update: A View Into AI Innovators

Two months ago, in partnership with Anthropic, we launched the $100 million Anthology Fund, designed to accelerate the creation of novel AI-first solutions built with Anthropic’s powerful AI technology. The response was overwhelming. 

Since announcing the Anthology Fund, we’ve received thousands of applications from pioneering founders representing almost every continent. We’ve identified numerous investment opportunities from this pool of applicants and have written several checks already.

We’re keeping the details of our growing portfolio under wraps for now, but we saw an opportunity to answer some of the questions everyone is asking: What are people building with AI? Which sectors are attracting the most attention from startups? Our application data provides a unique glimpse into the emerging landscape of AI-powered companies. With Claude’s help, we’ve distilled thousands of ideas into insights that reveal where founders are focused. Here’s what we discovered about the next wave of AI-powered companies.

The Applicants: Tech Experts With Diverse Strengths

We were excited by the highly technical nature of our applicant pool—80% of our applicants came from strong technical backgrounds.

But “technical expertise” is one of the few commonalities. Each founder brings a combination of unique strengths and skills that set them apart, from deep domain expertise to entrepreneurial grit and hustle. Claude’s analysis surfaced a fascinating array of founder profiles—some blending technical prowess with entrepreneurial talent, while others stand out for specialized skills in areas like quantum computing, advanced statistical modeling, and complex systems analysis. We received a large number of applications from multi-faceted innovators, numerous founders identifying as luxury disruptors, several gritty self-starters converging around edtech, and dozens of quantitative savants—all eager to drive change with AI.

Innovation Across Every Category

Not surprisingly, a diversity of talent results in a diversity of ideas. We’ve received proposals for AI-driven diagnostic tools, platforms transforming personalized learning, and solutions aimed at supply chain optimization, to name a few. Founders are bringing AI into sectors as varied as transportation, healthcare, fintech, gaming, coding, and cybersecurity. The range of industries and sectors reflects the potential of AI to impact every aspect of our lives

Pre-Product to Early PMF

The Anthology Fund backs companies from pre-seed to Series A, so it’s no surprise that most applicants are pre-product. However, some startups already show signs of traction. A subset of businesses have achieved proof-of-concept milestones, built MVPs, or can show momentum across users, revenue, and growth, signaling strong product-market fit early on. Their early traction is a promising indicator of what’s to come.

The Road Ahead

The first two months of the Anthology Fund have surfaced an exciting cohort of founders championing ideas that will leverage AI in ways we’ve only begun to grasp. It’s inspiring to see what the best builders in the world can do when given powerful tools like Claude 3.5 Sonnet to build with. 

But the Anthology Fund is about more than just financial backing. We’re fostering a community of innovators at the forefront of the AI revolution. Startups supported by the fund will have access to Anthropic’s cutting-edge technology, $25,000 in credits for advanced AI models, and the deep company-building expertise of Menlo Ventures. (You can learn more about these benefits and submit your startup application here.)

And this is only the beginning. Stay tuned as we continue this journey with our partners at Anthropic and the visionary founders shaping the future.

Together, we will ignite innovation and help create the next generation of game-changing tech companies.

Methodology

To analyze thousands of text descriptions effectively, we developed a multi-step process that combines machine-learning techniques with human-guided interpretation:

1. Data Preprocessing: We applied basic filtering to the dataset to remove spammy entries or submissions with too short a description and maintain a manageable sample size.
2. Text Embedding: Using a sentence transformer model, we converted each text description into a numerical representation. Specifically, we relied on HuggingFace’s “all-MiniLM-L6-v2” model, which balances performance and processing speed for short paragraphs. This step resulted in 384-dimensional vector embeddings for each description.
3. Dimensionality Reduction: To visualize the high-dimensional data, we applied t-SNE, (t-distributed Stochastic Neighbor Embedding). This technique projects the 384-dimensional vectors onto a 2D or 3D space while aiming to preserve the relative distances between points.
4. Clustering: We then grouped similar descriptions using the K-Means clustering algorithm and iteratively adjusted the number of clusters to find the most meaningful groupings.
5. Cluster Labeling: To make the clusters interpretable, we utilized the Claude 3.5 Sonnet AI model to generate concise, one-line summaries for each cluster based on the descriptions it contained. These summaries were then plotted alongside the cluster visualizations.

This approach allows us to efficiently organize and interpret large volumes of text data, providing insights that might be challenging to obtain through manual analysis or simple keyword searches.