Organizational development, from the ground up
Tacit Inquiry is where we start. It isn't the whole job. Organizational development asks how the whole team works with AI, and answering that takes ten kinds of work.
Most AI use still happens in one person's chat window. What that person learns stays in their chat history, so the team's AI starts empty. Research on AI at work keeps finding the same thing: individual gains don't add up to organizational gains without changes to how the organization works.
Small organizations feel this most. Between December 2025 and May 2026, AI use rose among U.S. firms with at least 20 employees, but didn't change significantly among smaller ones. Only 39% of people who use AI at work say their company has trained them.
1x1 conversations with the people who do the work, reviewed by them: the workarounds, handoffs and judgment calls nobody has written down.
Knowledge lives in four places: the systems, the written rules, the workflows as people practice them, and the unwritten norms. Access to the systems only reaches the first.
Task by task, decide what AI can take on, what it can support, and what it should leave alone. Automation comes last, because automating an undocumented workflow automates the guesswork too.
Who decides, who checks, and when work hands back from AI to a person. Designed with the people who do the work, not for them.
What AI can read and what it never should: personnel files, student records, donor and financial data. Proper access matters more than wide access.
Hands-on training and coaching so the whole team benefits, not only the early adopters. That includes knowing when not to use AI.
Move what people learn out of one person's chat history and into the team's knowledge vault, reviewed and owned. From me, to us, to all of us.
People need to be able to say what they don't know, admit they use AI, and say when it's wrong. Staff should know what's read, why, and who sees it.
Ground-up work needs a sponsor at the top. Leaders set the rules, use the tools themselves, and protect the time it takes to listen.
Read real AI answers, note what's wrong in plain words, and fix the knowledge, not just the prompt. A knowledge vault only fills through listening and review.
For one AI workflow, list what it can read and what it has to guess.
With one person within five years of retirement, about how they actually do the work.
From your AI last week. Note what's wrong in plain words.