Thought leadership

Notes from the middle of it.

Working notes on where AI capability, cost and adoption actually are — written for people who have to make decisions with them. Every figure is redrawn from published data and carries a table view, and where a source contradicts itself we say so on the chart rather than quietly picking a number.

Published pieces

  1. Interactive briefingOctober 20263 min executive view

    The enterprise becomes programmable.

    AI capability and cost moved further in twelve months than most operating models did. The constraint is no longer access to intelligence; it is how the enterprise is designed to use it. Interactive, fully sourced and built to present.

    1. 8shiftsin what AI can do and what it costs
    2. 6decisionsfor the leadership team, each with a suggested owner
    3. 90daysan agenda to assign, then copy as a brief
    Open the briefing

    Or start in a reading mode

  2. Technical BriefAugust 20269 min read

    Frontier capability stopped being the bottleneck.

    The price of frontier-class intelligence fell by roughly two orders of magnitude in thirty days, while the five leading models converged to within nine index points of each other.

    • 137× spread in cost per task across the board
    • More reasoning effort is a cost multiplier, not a quality dial
    • 84% of the world has never used a model at all
    Read the brief
  3. Quarterly briefingJuly 202612 min read

    Adoption is universal. Impact is not.

    Almost everyone is using AI and a minority can find it in the P&L. The separation comes from redesigning the operating model, running agents under real controls, owning the intelligence layer, and building rather than renting.

    • 88% use AI regularly; 39% see enterprise EBIT impact
    • Under 12% of companies have AI-ready data
    • The operating layer is the only tier where advantage compounds
    Read the brief
  4. Quarterly briefingApril 20269 min read

    Stop treating AI like a smart intern.

    It stopped being a sidecar for note-taking and became an execution layer. What separates the next twelve months is cycle time, execution quality and the distance between a decision and the work that follows it.

    • A 30-day operating review beats a 6-month pilot
    • Anything built a year ago is almost certainly overspending
    • Five moves, each with an owner, a clock and a target
    Read the brief
  5. Annual outlookJanuary 202611 min read

    The model is not the moat.

    Access to a frontier model stopped being a differentiator the moment every serious competitor had the same access. What separates companies is the layer above it.

    • The moat moved to the context layer
    • Managed autonomy, not full autonomy, is what is working
    • Software increasingly gets built rather than bought
    Read the brief