We embedded 24,000+ companies into 384-dimensional vector space using all-MiniLM-L6-v2. Then we looked at what the geometry tells us about industry boundaries, agent readiness, and the invisible threads connecting AI companies.
Each dot is a company pair. X-axis is their cosine similarity (how alike their descriptions are). Y-axis is the gap in their Agent Readiness scores. If similarity predicted AR, we'd see a downward slope. Instead, the story is more interesting.
Hover over dots to see company pairs. Data from 5 reference companies x 20 neighbors each.
Spider charts show the "shape" of each AI sector across five dimensions. Agent Frameworks are high on AR but low on B2B; Healthcare AI is the reverse. These shapes explain why similarity scores cluster the way they do.
A force-directed graph where companies are nodes and edges represent similarity > 0.6. Thicker edges = higher similarity. Color = sector. Watch how clusters form naturally from vector space geometry — no human labeling needed.
Each cell is a company. Color intensity = AR score. Grouped by sector. The Infrastructure & MCP sector consistently scores highest, while Healthcare companies cluster at the bottom despite high similarity within their group.
Semantic similarity is not AR similarity. Companies that describe themselves in nearly identical language (cosine > 0.8) can have AR gaps of 50+ points. This means:
This is why OnlyData scans actual infrastructure (llms.txt, MCP servers, OpenAPI specs) rather than relying on marketing copy.
All 228 AI Economy companies with similarity scores, AR scores, and full profiles.
Browse AI Economy Dataset