The studio
Cold Anvil Studios is a one-person AI research and development operation. It builds specialist harnesses for consumer and business use, on research that identifies the highest-performing combinations of harness, model, provider and treatment (skills, plugins and the like). The goal is products that get the most out of the models available today: better bang for the buck than general-purpose frontier tools like Claude Code.
Current projects
All projects
Active
JackBench
An evidence-driven evaluation system for measuring the complete model, provider and harness combination rather than the model alone. It compares frontier and near-frontier models across coding-agent harnesses, with preserved traces, blinded judging and publication gates.
In Development
The Forge
The harness that runs the studio's agents. Plans get checked independently before any work starts, finished work has to clear two verification bars, and everything in between is written to an audit log. The intent is to open-source it once it's been evaluated properly.
Concept
Beararms.ai
Nothing public. Yet...