Writing
Long-form analysis for leaders building regulator-defensible architecture, AI governance, evidence, portability, and durable modernization.
New and noteworthy
The three most recently published essays, ordered by publication date.
Seven practical principles for building and scaling a Data & AI organisation: start with the business mandate, hire for judgement, develop leaders early, make accountability explicit, use governance to accelerate delivery, design for multicultural work, and measure organisational value.
Swiss employers are pricing tools, locations, and engineering seats while underpricing the institutional context, specification work, and accountability that regulated data delivery requires.
Why AI lifecycle evidence must be captured while systems are designed, tested, approved, released, and monitored—before models, prompts, suppliers, and people move on.
Browse by topic
Explore connected essays through canonical expertise briefings rather than a wall of overlapping tags.
View all topics →AI governance helps companies use AI safely by defining ownership, controls, metrics, and review paths.
AI architecture connects business goals with data, model, governance, and platform decisions.
An AI SDLC makes AI delivery testable, governable, observable, and maintainable.
Strong data engineering is the foundation for credible AI, analytics, and operational decision-making.
Engineering management turns technical ambition into repeatable delivery and healthier teams.
Human-AI collaboration is about workflow design, not just tool access.
Archive
An AI-ready engineering organization can attribute, verify, reconstruct, and defend every AI-assisted change through evidence, controls, and accountable ownership.
AI has made software generation cheaper, but verification, evidence, security, accountability, and regulatory defensibility remain the real enterprise cost.
Agent output can scale much faster than the reviewed, defensible judgement available to accept it. Safe team design starts by pricing that supervisory capacity.
AI transformation creates durable value when organisations redesign workflows, roles, learning, accountability, disclosure, and human oversight—not when they merely increase licence utilisation and prompt counts.
The Fable 5 shutdown showed that frontier models can become externally controlled dependencies for regulated enterprises. That changes how firms should think about sovereignty, portability, concentration risk, and operational resilience.
Frontier AI models are global, but government and regulated data remain jurisdictional. This article explains why sovereign AI is becoming the default architecture across Europe.
Most governments now have AI strategies, principles, and playbooks. The harder question is whether they have the delivery machinery to turn those documents into safe production systems.
Part 3 of the AI software engineering series: why AI adoption becomes an operating-model problem, what the emerging AI engineering stack looks like, and what leaders should ask Monday morning.
Part 2 of the AI software engineering series: AI exposes existing enterprise weaknesses, changes requirements, raises the cost of architecture mistakes, turns testing into evaluation, and moves governance into runtime.
Part 1 of a three-part field guide on how AI changes software engineering: code generation accelerates, but the bottleneck moves toward architecture, evaluation, governance, ownership, and judgment.
Why enterprise AI adoption often stalls between promising pilots and durable production: the bottleneck is usually data, governance, ownership, workflow integration, and operating model maturity rather than model quality alone.
From warehouse to lakehouse to DataOS: an architectural genealogy of enterprise data systems and why the next generation must solve sovereignty, evidence, portability, and AI governance together.
An introduction to the Sovereign Data Operating Plane (SDOP): a regulator-defensible architectural pattern composed of five elements and three differentiating pillars.
Why Tier-1 enterprises are simultaneously struggling with enterprise AI, regulator-defensible data architecture, and continuous modernization, and why current architectures solve at most one of the three.
Eleven years after BCBS 239, only two of thirty-one G-SIBs are considered fully compliant. This article examines what that failure reveals about modern enterprise data architecture.
