What we do
We work with AI systems daily. Running local infrastructure, building agentic workflows, testing frontier models, documenting what we learn. The research comes from doing.
We are interested in how AI actually behaves when you rely on it for consequential work. Not benchmarks. Not marketing. Real use.
Why this exists
Most organisations using AI have no independent way to assess whether their reliance is reasonable. They understand what systems produce; they rarely understand how outputs are being relied upon, or where accountability has quietly shifted.
We built this lab to understand AI from the inside. Not to sell services, but because these questions matter and someone should be working on them seriously.
Approach
We build and test. Local infrastructure means we can run experiments that would be impractical through API access. Daily use of frontier tools means our observations are grounded in practice.
When we find something interesting, we write about it. Our work is open by default.
Team
Founder
High Court Enforcement Officer
One of 48 High Court Enforcement Officers appointed by the Lord Chancellor. A statutory role involving High Court writ execution and consequential decisions under legal accountability. Also deeply immersed in frontier AI: local infrastructure, agentic workflows, daily use of tools like Claude Code. This combination of institutional credibility and hands-on technical work shapes everything we do.
Infrastructure
We maintain dedicated computing infrastructure for AI research, including NVIDIA Blackwell, DGX systems, and multi-GPU configurations. This allows us to run local language models under controlled conditions, test across model versions, and conduct experiments that would be impractical or impossible through API access alone.
Local infrastructure matters for research integrity: complete control over parameters, no rate limits during intensive testing, ability to reproduce results precisely, and freedom to publish without commercial restrictions.
Philosophy
We believe in evidence over opinion, documentation over assertion, and clarity over hedging. We do not hype and we do not hedge. We document what we observe and explain what it means.