When we began scaling GenAI, I described the challenge as a duality: risk on one side, opportunity on the other, and leadership as the act of navigating between them. Most organisations I’d worked in resolved that tension with control: heavy-handed access restrictions, long-running policy programs, rigid governance. Well-meaning, and blocking momentum. So we flipped the script. We engaged the curious, treated early low-risk missteps as learning rather than failure, and invited our people to co-design the rollout. At the time, I wondered aloud whether trust could be the bridge between risk and opportunity.
Having now lived the experiment, I can report back: I had the geometry wrong. Risk and opportunity were never two banks of a river needing a bridge. Almost everything we did to manage the risk turned out to be the same work that unlocked the value.
Look at what actually happened. The early missteps surfaced real exposures, in places no policy workshop would have predicted, and fixing them made us better. The people we asked to interrogate outputs and check sources became our most credible educators. Every process that needed a named owner before an agent could touch it, every dataset that had to be cleaned, every access rule that had to be sharpened: risk management on paper, but capability building in reality. The organisation that is safe to scale AI and the organisation that gets value from AI are the same organisation. You build both or neither.
I’d also correct something about my early thinking. I talked about trust as though it were a sentiment. It isn’t. Trust without architecture is just hope. What made trust scale was structure: education before access, clear zones so people know where their judgement is the control, named owners, and an expectation of honest reporting when things go wrong.
So stop asking how to balance risk and opportunity. Ask instead what would make AI safe to scale in your organisation, and notice, as you work through the answers, that you are simultaneously building the thing that captures the value. The duality dissolves. It was one job all along.