Don’t make your workforce wait for AI

An open letter to CIOs.

A meeting that was supposed to take thirty minutes ran fifty. Not because the agenda was heavy — because the CIO on the call couldn't stop pontificating about how his org would "figure it out" once the AI rollout landed. No plan, no timeline, no owner. Just a vague faith that competence would materialize on contact with the tool. I've sat in enough of these meetings across enough platform shifts to know exactly how that story ends: it doesn't. Not on its own.

So this is an open letter, because I keep having some version of that same conversation, and I'd rather say it once, in writing, than keep having it in fifty-minute increments.

Here's the thing about AI at work that most rollout plans miss: your people already have an opinion about it, and it was formed at home. They're using it to plan trips, draft the awkward email to their landlord, help their kid with algebra. They know what fast and useful feels like. So when they get to work and the "AI transformation" turns out to be a locked-down chatbot with a six-week approval process for a use case someone in IT hasn't thought of yet, the gap isn't subtle. It's the difference between a tool and a permission slip, and people can feel it immediately.

That gap is the real risk, more than any of the ones on your security review. If the tool at work is worse than the tool at home, adoption doesn't fail loudly — it just quietly doesn't happen. People route around it. They keep using the version on their phone, off the record, and now you've got shadow AI instead of sanctioned AI, which is a strictly worse position than the one you started in.

What actually works, from what I've watched play out, isn't a single big-bang platform launch. It's closer to a parallel path: give people something usable now, even if it's not the final-state architecture, while the enterprise-grade version gets built properly in the background. The instinct to wait until it's perfect and secure before anyone touches it is understandable and also exactly backwards — you're optimizing for the audit, not for the adoption curve, and the adoption curve is the thing that actually determines whether any of this pays off.

The ROI case, if you need one for the board, isn't complicated. Give a knowledge worker back even a few hours a week and the math works itself out well before the license renewal does. I won't bore you with a spreadsheet — you already know how to build that model, and honestly the harder number to estimate isn't the time saved, it's the cost of the alternative: a workforce that's already fluent in AI on their own time and increasingly unimpressed by what they're handed at work.

If I had to boil this down to one thing I'd want a CIO to actually act on, it's this: stop treating AI adoption as an IT rollout and start treating it as a change in what "normal" looks like for how work gets done. Rollouts have a launch date and a completion percentage. Norms take root because people saw someone they trust use the thing and get something real out of it. You can mandate the first. You cannot mandate the second — you can only make room for it.

The CIO from that fifty-minute call wasn't wrong that his org would eventually figure it out. Every org eventually does. The only real question is how much time, trust, and shadow-AI risk gets burned in the "eventually."

D.