Research-backed essays on AI adoption, change management, and how people actually work with new tools.
Adoption is effectively universal; measurable returns are not. A desk's-eye reading of the 2025–26 data from Stanford, McKinsey, MIT, Gartner, and Menlo Ventures — and what the gap between the two means if you sell AI work.
The practical shift from chatbots to real AI agents is not that work disappears. It is that more of the day becomes delegation, inspection, and recovery.
Claude Code is the most powerful AI coding tool most non-technical professionals have never heard of. Here's how to go from zero to shipping real software — no coding background required.
ADKAR was built for organizational change. GenAI adoption is organizational change. Yet most AI rollouts skip the Awareness and Knowledge stages entirely and wonder why Reinforcement never sticks. Here's a practical mapping.
Employees consistently overestimate their AI competence — by a factor of 5. I measured the Dunning-Kruger effect directly in a controlled workshop. Section AI's research confirms the pattern at scale: confident users are often the least prepared.
McKinsey's landmark statistic has defined the transformation conversation for a decade. But the reason for failure has always been people, not technology. Here's how generative AI creates both new risks and new opportunities for getting it right.
Organizations are deploying AI tools at scale while productivity metrics stay flat. My 6-month research at Utiligize uncovered why — and what it actually takes to translate individual AI gains into organizational results.
Short observations from real AI work — the things too small for an essay but too useful to lose. Updated often.
Six months ago the hard part of an AI task was getting a usable first draft. Now the models clear that bar easily, and the slow step is checking the output well enough to trust it.
Teams that feel "fast with AI" are usually just good at inspection: they have a cheap way to verify a result before it ships. The ones that feel stuck are re-reading everything by hand because nothing tells them where to look.
The intake brief that makes verification explicitA pilot succeeds because a motivated person babysat it. That tells you the tool can work — not that it will work once the babysitter moves on.
The useful question after a pilot is not "did it work?" but "what did the person do that the next ten people will not?" That gap is the real rollout plan.
A shared prompt library feels like leverage, but most entries are tuned to a model version and a task that has already drifted. Six weeks later half of them quietly underperform a plain request.
Treat prompts like code, not like quotes: date them, note the model, and delete the ones nobody can explain.