24 posts
Writing
Most of this is about the unglamorous middle of AI adoption: what happens after an organization decides to use AI, when people need room to try it safely and somebody has to make sure what they learn gets used.
Most of these were first published in November 2025 and revised in September 2026. Where the facts have moved since, the post says so.
Start here
If you read nothing else, these make the core argument, in this order.
Why AI pilots stall, and what to build instead
Most AI pilots stall for organizational reasons. A budget, a sandbox and a shared record of what works do more than another round of pilots.
Experimentation and adoption · 4 min read · Updated September 2026
An AI budget for every employee
Give people a small monthly AI budget, no approval per experiment, inside a sandbox. The people closest to the work will find the use cases.
Experimentation and adoption · 4 min read · Updated September 2026
Safe early access to new AI tools
The bigger risk is often the year spent deciding whether a tool is safe. A sandbox lets people try it in weeks, with the risk contained.
Experimentation and adoption · 4 min read · Updated September 2026
Why teams rebuild what already exists
Organizations pay for the same solution again and again because nobody can find the first one. AI agents will do it faster unless reuse gets easier.
Experimentation and adoption · 5 min read · Updated September 2026
Measuring AI when ROI doesn't fit
Payback-period ROI suits projects that replace a known process. A lot of AI isn't that. Here is what I'd track alongside it, in terms finance can audit.
Leadership and governance · 4 min read · Updated September 2026
By topic
All topicsExperimentation and adoption
Budgets, sandboxes and shared learning: how people actually start using AI at work.
Leadership and governance
Where decisions sit, how ideas get rewarded, how skills and roles change, and how to tell if it’s working.
Data and knowledge
Metadata, content lifecycle and the data plumbing that agents depend on.
Vendors and platforms
SaaS lock-in, data portability, Microsoft Copilot, multi-cloud and build versus buy.
Agents and tools
MCP, orchestration, coding agents, and a longer view of where this is heading.