Insights

Notes on building serious software with AI, and the discipline that makes it work.

• Whitepaper · PDF
  Architecting Agentic Systems for Accuracy and Fairness

  The architectural controls that make agentic AI accurate and fair — fairness rubrics, grounding and verification, memory governance, human oversight, and the EU AI Act / Consumer Duty anchor. Free download.

• The runtime problem: why your AI agents fail in production

  Most enterprise AI agents that fail do so not because of the model but the runtime. Why durable, queryable, temporally honest state is the missing primitive — and what it changes about how you architect.

• Securing an agentic solution

  An agent is software that chooses its own actions — and that breaks the assumptions traditional application security rests on. A build guide to the four boundaries, mapped to the OWASP Top 10 for Agentic Applications.

• Your collateral optimiser is only as good as the data beneath it

  Banks have spent heavily on collateral optimisation, yet much of it underperforms. The bottleneck is rarely the algorithm — it is the fragmented, point-in-time data layer beneath it.

• Graph databases: from graph theory to enterprise applications

  From Euler’s bridges of Königsberg to AI, fraud detection, payment routing and grid optimisation — and why, for connected problems, graph databases leave the relational model standing still.

• How Claude Got Its Name

  The man who measured information was called Claude Shannon. The history that runs from a 1928 question about proof to the machine that now bears his name.

• AI-first development is a discipline, not a tool

  How a small team can ship at the pace of a large one, and why most teams get this wrong.