Almost a year ago I wrote about the Goldilocks Dilemma. Centralised push versus agile pull. Control versus speed. I didn’t have an answer then. I said time would tell.
Time’s told.
I leaned into pull. Access for anyone who wanted it, encouragement to experiment, people sharing prompts and wins and failures with each other rather than waiting for a training calendar. It worked, mostly the way I hoped. The curious ones ran with it. GenAI stopped being a novelty and became part of how people actually work, the same way it did for me. Some of the best ideas in the organisation came from people nobody asked to be innovators.
But pull on its own has a blind spot, and I found it the hard way. The people who didn’t experiment stayed exactly where they were, and the risk didn’t distribute evenly either. A handful of enthusiastic early adopters can quietly create a handful of enthusiastic early risks — data going places it shouldn’t, decisions being made on outputs nobody checked. Agile is great until the thing that slips through the gap is something you can’t take back.
So the answer wasn’t pull instead of push. It was pull, with a net underneath it. We stood up a temporary AI Steering Committee — not to slow people down, but to catch the orphan risks that don’t belong to any existing committee yet. Somewhere for the edge cases to land while we work out which parts of “how we already manage risk” simply absorb AI, and which parts genuinely need something new. It’s deliberately temporary. The goal is to make itself redundant as governance catches up.
That’s the bit I got wrong a year ago, or at least didn’t see yet. I thought of push and pull as a dial you set once. It’s not. It’s push for the guardrails and pull for the learning, running at the same time, and you keep adjusting both as the organisation’s confidence and your risk appetite move.
If you’re earlier in this than I was, my advice hasn’t changed as much as you’d think: still lean toward pull, still trust people to learn by doing. Just don’t wait as long as I did to build the net.
So, a year on — how’s your Goldilocks Dilemma looking now?