Issue 01
Hiring AI leaders for the wrong problem
93.2% of data and AI leaders say the constraint is culture, change and organization. 6.8% say technology. Most VP-level specs are written for the 6.8%.
Newsletter
$725B of AI spend is public. The returns aren't. I close that gap for a living — this is the research.
Subscribe on LinkedIn →What this is
Every issue follows the same five sections: the gap, why it persists, what it costs, what to do Monday, and the receipt — sources, predictions, and what gets tracked next.
Predictions are stated plainly and checked in public. When one is wrong, it gets marked wrong.
Issues
Issue 01
93.2% of data and AI leaders say the constraint is culture, change and organization. 6.8% say technology. Most VP-level specs are written for the 6.8%.
Issue 02
Enterprise AI capability rose from 88% to 93%, but the share of companies whose AI returns beat their spend stayed stuck at 57% — because the AI lead is hired on a technical spec, so reports track movement (seats, use cases) instead of business outcomes, with no baseline to prove anything moved.
Issue 03
Hyperscaler AI capex hit roughly $725 billion for 2026, but the winners so far are the suppliers (Nvidia, TSMC, SK Hynix climbing the Fortune 500). Deployment, adoption, and return are three different tests, and most enterprises can only answer the first
Issue 04
Models got 95% cheaper, so intelligence is no longer the advantage — it moved into governed context: what the agent may know, which rule it applies, and what it's allowed to do next. She defines the Decision Spine as the six layers (data, quality, meaning, policy, workflow, evidence) connecting knowledge to defensible decisions. Here is a claude skill that you can run on your own to test any bot.
Artifact library
Decision Spine Exam
Protocol for testing public AI assistants against current company policies using observable evidence. Guides scenario setup, live testing, evidence scoring, and reporting.
Decision Spine Exam — ChatGPT
Decision Spine examiner for testing public AI assistants, with ChatGPT-specific guidance and standard prompts for testing a decision layer.