Companies

A hand-picked roster we read daily — what they shipped, what they are hiring for, what is breaking. Each one carries a line on why we track it. Everything below this band is coverage; this is the part we are in the room for.

• AgentaThe agent harness as a product — an open-source workspace that versions an agent's instructions, skills and tools like code and evaluates a change before it ships, which is the layer the whole "build your own system" argument runs through.
• AI ScoreThe governance and observability layer over enterprise inference — the system of record for what an organisation's AI and agents actually did.Seed
• BluedotA high-volume inference consumer paying retail — every meeting they capture is transcription plus summarisation billed per token, and their margin is the gap between a $14–$39 seat and what that seat costs them to serve.
• Cactus ComputeThe bottom of the inference market — a 14MB model and a dependency-free C++ runtime aimed at the four in five edge devices that cost under $200, where the unit economics are flash reads and milliwatt-hours rather than tokens.
• CallosumThe routing layer across heterogeneous silicon — the software that decides which model runs on which chip, sitting between the chip makers and the teams buying inference.Seed13 open
• CerbosThe runtime permission layer for AI agents — the company answering "what is this agent actually allowed to do", which is the control almost every agent stack is currently missing.Seed

41 more on the roster