In-depth analyses of engineering, security and governance, with traceable sources and practical conclusions.
A technical analysis of METR's measurements of AI-assisted software development, their limitations and the implications for credible internal evaluations.
The studies do not provide a universal productivity number for coding agents. They show how closely outcomes depend on tool maturity, task selection and measurement design.
Sources reviewed: 2026-07-18
Read the study →A safe reconstruction of the path documented by GitHub Security Lab and the technical controls required for agentic development environments.
The weakness does not reside in a single prompt. It emerges when untrusted content, powerful tools, local secrets and outbound communication are connected without end-to-end trust boundaries.
Sources reviewed: 2026-07-18
Read the study →An analysis of GOV.UK Chat's public Algorithmic Transparency Record and what it contributes to governance, architecture and accountable operation.
The record is valuable because it joins ownership, purpose, components, data flows, evaluation and retention into an inspectable operating picture. It documents governance practice but does not replace it.
Sources reviewed: 2026-07-18
Read the study →Briefly describe the situation, your objective and the systems involved. We will respond with a technical assessment and a practical next step.