Grant Ecker had just handed me a lovely image. The human side of technology change, he said, is like the butter on top of a steak you?re cooking: ?it’s still there. It’s just dissolved.? And I sat there on the call thinking, I’m not sure I’ve seen it melt.

Grant came on Whatever Next? Unplugged to talk about leading the architecture practices at Medtronic, Danaher, Walgreens Boots Alliance and now Ecolab, and about the Chief Architect Network, the volunteer community of around 725 architecture leaders that has built with the architect community. He has a global view very few of my guests bring. He also cares about the human side, out loud, which is exactly why I pushed back. Because if the human in the loop is going to matter in the age of AI, we need to be honest about how much of the human side ever really made it into the dish.

Embedded, or just not talked about?

Grants argument goes like this. We don’t really talk about our internet strategy any more. It’s just there. The same will be true of AI. He quoted Jeff Winter, from our network, on AI being ?significant in the business and in the impact, but invisible in the language?. And he put the human layer in the same place: so much a part of how technology gets done now that we’ve stopped naming it.

Now maybe that’s true for the technology. But the human side?

When we get together with our architecture communities, and I’ve been in those rooms for the last fifteen years, I haven’t heard as much passion for the human side as Grant was showing on our call. I asked Wendy the same question when she was on the show: in those community conversations, what percentage is really spent on the technology, and what percentage on the human factors?

So I asked Grant. And to his enormous credit, he didn’t defend the metaphor. ?Yeah, okay, I think that’s fair,? he said. ?I think I’m kind of picking the best examples that I think really brought humans and the technology together.?

That’s the honest bit. The stories we tell about our best programmes are the stories where the human side was there from the start. They’re the exceptions. Grant’s own exception was Connect, an internal Facebook he built at General Mills, where he and his boss flew round the world to get people excited about sharing what they knew. Far more often, by his own account, ?the human engagement was sort of the afterthought.?

What the afterthought costs

Grant described the pattern he grew up with. You build a workflow system, you gather the business fields, you work out who approves what next. That’s as close as you get to the people. Then you deploy it and someone actually has to use it, and the shock arrives: ?what do you mean they have questions? Like what do you mean this isn’t obvious for them right out of the gate??

He’s quick to say we can laugh at ourselves for that now. There are whole disciplines in usability and user experience. But, in his words, it’s not complete.? There are still plenty of systems where requirements go in through an open door and ?an ugly system with a million buttons and a million fields comes out the other side and the company has to deal with it.?

The company has to deal with it. That?s the cost, and it lands on the people who never got asked.

And AI raises the stakes. I remember the term end user computing, and shadow IT, with teams around the business building their own Access databases. Today everyone has ChatGPT or Claude on their phone. People without any technical background are writing code and launching sites. Grant’s warning is that if the system we build doesn’t meet what people actually need, they’ll simply build their own, and we end up with ?a million one off solutions that don’t interrelate, don’t collaborate?.

If you lead architecture, or you sit on the board that funds it, ask yourself honestly where your last programme spent its time. On the technology, or on the people who would have to live with it?

Less about adoption, more about confidence

This is where the conversation turned for me.

With the transactional systems and the workflows, the challenge was adoption: buy in and change. When you’re looking at AI, I’m not sure that’s the right question. Is AI about adoption, or is it about building confidence and trust in what it?s producing?

The human in the loop is how you build that trust. As soon as people are in the loop, seeing the AI produce the right outcomes, they can move up the food chain and let the tool run on its own, because they’ve developed the confidence in it. So maybe it’s less about adoption, more about confidence, and about getting the guide rails right.

Grant took that and ran with it. He described teaching a child to ride a bike: training wheels first, then off, but you stay on the driveway for the first few laps, and you’re still there when someone skins a knee. ?So I think it’s about gradual delegation,? he said. As confidence goes up and risk comes down, you let more decisions go. ?It’s not that we ever fully exit. Exceptions always get escalated to the human. We just have less and less of them. And also those escalations become moments to learn.?

His example came from Mike Carroll, another member of our network, and his work at Georgia Pacific. The AI optimised a weeks truck loads inside a rule that said never pay a late fee. Then it spotted that one $75 late fee, on one case of Dixie Cups, would lift the whole weeks fill rate. So it asked a person. The person said yes, made the call to the customer, and the AI learned something about when that rule can bend.

That’s the human in the loop doing two jobs at once: checking the machine, and teaching it.

Start with one decision

None of this happens in weeks. There are pioneers already trusting these tools to get outcomes that are consistently right, and there’s a whole load of people coming behind. It’s a generational thing, an era thing, three, four, five, ten years. Grant agreed it runs over quarters and years, ?but it starts now. And if someone was able to do it in 2023, we can certainly do it in 2026.?

So on Monday, pick one decision your teams make every week and look at it the way Grant does. Where confidence is high and risk is low, let it go. Where risk is high or confidence is low, keep a person in the loop, and treat every exception they review as a lesson, for the machine and for them.

Because there’s machine learning and there’s human learning. If we want to close The Human Gap, the key is that as humans we keep learning, just as the machines do. We keep thriving and changing alongside them.

We don’t just step back and let the machines take over.

Whatever next?

You can hear the full conversation with Grant Ecker on Whatever Next? Unplugged.