What I Would Tell a Team Adopting AI Tomorrow (Learned the Slow Way)
Skip the training day and the policy doc. Five things I would tell any team starting their AI adoption, in the order I would say them.
A friend's company is starting their official AI adoption next month and asked what I would tell them. My answer is shorter than their rollout plan, and every item on it was learned expensively.
First: adoption is individual before it is organizational. The training day teaches features; it does not change habits. What works is finding the people already quietly using these tools, making them visible, and letting their workflows spread. Mandated tools with mandated uses produce compliance theater and a dashboard nobody believes.
Second: write the two-sentence policy before the fifty-page one. What data never goes into these tools, and what gets human review before it leaves the building. Everything else is discoverable. The long policy docs I have seen mostly regulate the past, written by people managing risks they have not yet taken.
Third: measure hours back, not usage. Dashboards full of prompt counts measure enthusiasm, and enthusiasm fades by month two. The metric that survives is boring: what recurring task now takes less time than it did, and can we see that in the work? Teams that track this build their tool decisions on evidence instead of novelty.
Fourth: budget for verification from day one. Every AI-assisted output that leaves the team carries the reviewer's name with it. The teams that skip this meet their first confident hallucination in public. The review hour is not overhead; it is the product.
Fifth, and it is the one they will not want to hear: some of what you buy will be unused by spring, and that is fine. Kill it at the audit, keep what compounded, and do not let the failures discredit the wins. Every good stack I know is three survivors wearing the grave dirt of a dozen canceled subscriptions.