Writing

Notes from the field.

Essays on enterprise analytics, applied AI, healthcare technology, public-sector data and the practice of leading technical teams. Written from inside the work, not from the conference circuit. Thirty essays, January 2024 to today.

2026

  1. May 2026

    The next five years of applied AI: a practitioner's view

    AI

    Five-year predictions are usually wrong. Here is a set anyway, written from inside the work, with confidence calibrated where I have it and uncertainty where I do not.

  2. Apr 2026

    Transformational leadership inside technical organisations

    Leadership

    Transformational leadership is a phrase that has been laundered by consulting decks. The actual practice is harder, more local, and less cinematic.

  3. Mar 2026

    When to fire your data warehouse

    Data

    Data warehouse migrations are usually wrong. The cases where they are right are specific and worth knowing.

  4. Mar 2026

    Building Aevora: what skilled home care taught me about software

    Healthcare

    I co-founded a home care business. Building software inside an operationally regulated business is different from building software for one. Here is what changed.

  5. Feb 2026

    Why your AI strategy keeps failing the board

    Strategy

    The strategy decks are good. The execution is uneven. The board is asking the right questions and getting answers in the wrong unit.

  6. Jan 2026

    The state of analytics engineering, 2026

    Practice

    The discipline has matured. The job has changed. A look at where analytics engineering actually is, away from the conference circuit.

2025

  1. Dec 2025

    Anti-patterns in multi-agent system design

    AI

    After a year of multi-agent system reviews, the failure modes have become repetitive enough to name.

  2. Nov 2025

    The CFO questions about AI you should be ready to answer

    Strategy

    Engineering teams are usually ready for the CTO's questions and unprepared for the CFO's. The second set is harder.

  3. Oct 2025

    Why I stopped reaching for vector databases by default

    AI

    Vector search is a real tool. It is also overused. For a lot of retrieval problems, the right answer is something simpler.

  4. Sep 2025

    What healthcare regulators actually want from your AI

    Healthcare

    The regulatory ask in healthcare AI is more pragmatic than most teams expect. Reading the guidance carefully changes what you build.

  5. Aug 2025

    The right way to think about AI ROI

    Strategy

    Most AI ROI conversations are conducted in the wrong unit. The conversation gets a lot easier when you switch to the right one.

  6. Jul 2025

    Document intelligence at scale: lessons from 37,000 lines of code

    AI

    I built and shipped a document intelligence platform that has now been licensed to a Fortune 1000 broker. Here is what I would do differently if I started again.

  7. Jun 2025

    How to read a data team roadmap and tell if it will deliver

    Leadership

    I read a lot of data team roadmaps. The good ones share four properties. The bad ones share four others. Both sets are worth knowing.

  8. May 2025

    The economics of self-hosted LLMs in 2025

    AI

    The math has shifted. Self-hosting is now defensible for a wider set of workloads than it was a year ago, but the cases where it wins are still narrower than the marketing suggests.

  9. Apr 2025

    Predictive readmission models that actually work

    Healthcare

    Most readmission models are graded against the wrong outcome. Building one that changes clinical behaviour is a different exercise.

  10. Mar 2025

    Government data modernisation needs more boring engineering

    Government

    The reason public-sector data work is hard has very little to do with technology and almost everything to do with constraints we are taught not to respect.

  11. Feb 2025

    Agentic AI is over-hyped. One use case is real.

    AI

    Most agent demos are toys. The exception is workflows where the cost of being wrong is bounded and the cost of being slow is high.

  12. Jan 2025

    The four failure modes of analytics teams

    Leadership

    Most analytics teams fail in one of four ways. Knowing which one is yours is the first useful step.

2024

  1. Dec 2024

    What enterprise AI got right, and wrong, in 2024

    Strategy

    A working list, with the benefit of being inside the engagements rather than reading the analyst reports.

  2. Nov 2024

    Building a semantic layer that survives a reorg

    Data

    A semantic layer is supposed to be a stable contract between the warehouse and everything that consumes it. Most are not stable enough to survive their first reorganisation.

  3. Oct 2024

    Notes from shipping AI to a Fortune 100 customer

    Practice

    What changes when the buyer is the size of a small country, and what most vendors get wrong about selling AI into that environment.

  4. Sep 2024

    Why generic LLMs underperform in clinical workflows

    Healthcare

    A frontier model is impressive on a benchmark. In a clinical workflow, the same model is fragile in ways that benchmarks do not measure.

  5. Aug 2024

    dbt or Airflow: a working answer for analytics teams

    Data

    The choice is not which tool is better. It is which work belongs in which tool, and where the boundary should sit. Most teams get the boundary wrong.

  6. Jul 2024

    Guardrails for LLM systems in regulated industries

    AI

    What guardrails actually mean in production, where the legal team gets a vote, and where the cost of being wrong is not measured in user satisfaction.

  7. Jun 2024

    The hidden tax of dashboard sprawl

    Practice

    Every dashboard you ship is a permanent commitment. Most analytics teams are running thousands of permanent commitments without anyone counting.

  8. May 2024

    How I rank technical debt as an analytics leader

    Leadership

    Not all debt is equal. Some pays interest. Some compounds. A small amount earns the team money. Knowing which is which is the leadership skill.

  9. Apr 2024

    Designing analytics for HIPAA without losing speed

    Healthcare

    HIPAA is not the reason your healthcare analytics team is slow. It is the reason your team is slow if you treat compliance as a phase instead of an architecture.

  10. Mar 2024

    RAG is a tactic, not a strategy

    AI

    Retrieval-augmented generation has become the default architecture for enterprise LLM work. That is a problem when nobody can articulate the underlying question.

  11. Feb 2024

    The data contract pattern that changed how I run analytics teams

    Data

    An interface between teams that produce data and teams that consume it. Boring on the surface. Career-changing in practice.

  12. Jan 2024

    Why most enterprise AI initiatives stall in pilot

    Strategy

    Pilots are not where AI dies. They are where the question of whether anyone wanted it in the first place gets answered, badly.