Skip to content
Stop Token

Blog.

  • · en · 7 min read

    Fifteen layers of your own AI: what each one does and what it costs

    The cost of the fifteen layers of a company AI assistant, a three-stage rollout, and open source, hybrid and SaaS compared over five years for three company sizes.

    Read →

    Fifteen layers of your own AI as a pyramid, with what each one does, what breaks without it and what it costs.
  • · en · 14 min read

    Centralize meaning, not data

    Why the core of an agent-ready data platform is the control plane, not the warehouse. Query the lake first, let usage decide what to define and materialize, and centralize meaning and control.

    Read →

    Centralize meaning, not data. Yesterday, meaning was spread across dashboards, a BI semantic layer and warehouse layers. Today, with agents: BI tools and AI agents ask through one policy model; a control plane holds canonical definitions, catalog, policies and placement rules; execution applies the same rules on every path; data lives in one open lakehouse copy; every answer carries evidence, and gaps and usage feed back into definitions and placement.Centralize meaning, not data. Yesterday, meaning was spread across dashboards, a BI semantic layer and warehouse layers. Today, with agents: BI tools and AI agents ask through one policy model; a control plane holds canonical definitions, catalog, policies and placement rules; execution applies the same rules on every path; data lives in one open lakehouse copy; every answer carries evidence, and gaps and usage feed back into definitions and placement.
  • · en · 5 min read

    Product builders: no results yet

    Five pilots at Meta, Atlassian, Intercom, epilot and Uber. More code gets written, little of it reaches users, and nobody has measured the business effect yet.

    Read →

    Meme: “We're product builders now.” “So you ship more features, or ship them faster, right?” Silence. “…right?”
  • · en · 7 min read

    The AI metrics tree: an attempt to measure the effect

    Net added hours and adopted increments at the top; adoption, quality, certified skills, AI-ready data and AI operating cost underneath. A raw tree, updated after the LinkedIn discussion, with what is not solved yet.

    Read →

    The AI metrics tree
  • · en · 7 min read

    Which AI words are rising, which are fading

    Thirty words of the English AI conversation, derived from Hacker News, GitHub, Wikipedia, OpenAlex and SEC filings: when each appeared, when it peaked, whether anyone still says it.

    12k
    comments on “ai-generated” in a year
    7 of 30
    words at their peak now
    79
    words tracked since 2022

    Read →data updated 2026-10-17

    Chapter 1 · the thirty words of the year

    AI-generatedat the peak
    training dataat the peak
    vibe codingbase
    AI agentsbase
    coding agentsbase
    AI toolsbase
    AI codingbase
    natural languagebase
    and 22 more words in the piece20222023202420252026
  • · en · 6 min read

    LAYRS: how a map of Paris became a marathon of cities

    One open-source 3D map of a city's buildings, a dozen cities rebuilt in a week, and what that says about building things now. With every link: ours, the Paris original, the community maps.

    Read →

    LAYRS: Amsterdam, 192,000 buildings by age
  • · en · 8 min read

    Data / AI Governance Program Map 3.0, with a self-consulting agent

    A map of Data and AI Governance on one Miro board, 120+ topics and 20+ exercises, rebuilt around what AI changed, with an agent that builds, audits or consults on a governance strategy.

    Read →

    Data / AI Governance Program Map 3.0