I have now been privileged to be leading three transformations in two companies. One of the three transformations is ongoing right now. And I'm seeing a few others as well on the side. Here's what doesn't work. So you can do more of the stuff that works.
The agent idea workshop
The first one that doesn't work, plainly, is to get a workshop going, collect agent ideas, and then put them in a grid of easy and hard and high impact, low impact. You know the classic: let's list the ideas, let's choose the best one, and then somehow magically the best one gets going.
You get the people together, you have a meeting. And what happens is that people will come to the meeting, they will have ideas, and then ideas get discussed. And then all of a sudden you have 20 ideas. Well, maybe 40. It doesn't matter. There's more than 10. There's more than 5. And they're on the wall.
You end up with logical ideas that are well defended.
You make a PowerPoint. Show the PowerPoint to higher management.
The more you have, the more you have to discuss and prioritize. All the effort goes into this social conversation on what's valuable, what's less valuable, what's easy and what's hard.
As you're starting with AI and agents and the transformation in general, you just simply don't know what is easy and hard. And most of the time, businesses don't know what's valuable and what's not valuable. Many of the ideas are meh. Some ideas might be great. But you just don't know which one will really fly.
The process may take a month. I've never seen anybody do this in a week in a real organization, because this involves people. Getting people really aligned usually just takes time and repetition.
You can claim that this selection is the first step in getting things going. In my humble opinion, overdoing this step is one of the worst stalls in making your transformation work. You are just stuck in analysis of the abstract.
The "What is AI native?" debate
My second gripe is the conversation and mulling over what is AI native.
This human want to define the target state and the target culture comes in many flavors. What should the teams be like? Do we need a backlog? What's the management system? What are the personal competencies that we need to have? What is the culture, and how do we know this? There are levels and maturities.
And the same effect: you get people into the room and people have a conversation. You write things down, there's material, the material needs to be discussed. You have opinions, somebody thinks otherwise, and it just balloons.
I'm starting to see the similarity: trying to think and speak your way into the transformation. The problem just compounds when there are many viewpoints to merge into a shared statement.
Anything first
The third is anything first, and then making it into a prerequisite for action: we can't do much before we have this and that handled. Call it "anything as a hard first dependency".
The can't-do-anything-yet comes in many, many forms. We can't do anything before we have firmly selected the right LLM platform, or the LLM vendor to be used by all. Or we cannot do anything, and we shouldn't do anything, before we have the necessary governance in place. We shouldn't do AI before our process is rethought. We think it is either too risky or inefficient. Sometimes both.
If the first two were meta discussions, now we are in the second-level rabbit hole: a meta discussion on what are the right foundations to do anything. What's the right foundation to make our company AI native? Ugh. If the first use case selection is hard, or the AI native discussion is hard, this meta-level is, again, twice as hard.
Again: more thinking and talking before doing. More analysis before learning.
Some companies know that they cannot analyze their way into the right selection. They take an educated guess. They simply throw a dart at the board, knowing that it will not land right, but whatever lands, they act on it. I suppose that's the right way. That might be the only way.
What works
Now, it's very logical to run the use case workshop. But it always seems to fail with the "what next". Because the what next is the hard part. After a use case selection, what is hard is actually to get people taking the time and using the tools, understanding what happened and what was learned, and to rinse and repeat.
The first attempts are flawed and broken. You literally don't know what you are doing. You learn. And that learning, as we notice, is the core.
I have now, in a way, lived through three such cases. All of them have been in a mode where we don't know, and I don't know, what is the absolute right action. And still, the feeling has been that acting is the thing that teaches us what we really need to know. That is the first pillar.
The second pillar is: "everyone has to act". It's a deliberate choice that this acting on AI is something that everybody has to do. It's not only management, it's not only team leaders, it's not only the seniors or the architects that need to kind of figure it out first. No, the other way around: everybody needs to act on something. And that action can be pretty much anything with AI.
That action creates the set of outcomes, the set of learnings. Half of the early creations will be right and half of them will be wrong, and there are gradients to rightness and wrongness. Half of the learning is false starts. But if you really think about it, you have created a huge set of knowledge within a month, as compared to mulling over the first use case to maybe try when you have time.
And then what happens is that people get going.
Get people going. Compound the skills, literacy, and wins. Most importantly: then you see.
Why it works
Why this works is something I was kind of expecting, but I didn't realize the situation on the ground is so strong.
By now, I would claim that 90% of people know that agentic AI will be a transformational thing. People simply expect that power to be there. That is the same as awareness. And that is also very close to desirability. People desire to actually be on the boat, to be part of it.
And people have been waiting. They have been waiting for the platform. They have been waiting for the use case. Anybody and everybody has been waiting for the kind of freedom to act, and the freedom to act on something that they believe they have a chance at achieving.
If you enable action by everybody and simply set people free, all of a sudden you have a lot of movement. And as the management saying goes, you can't steer what's not moving. You steer something that is moving.
P.S. The article is written with absolute claims about those things not getting people moving. I think smart managers will notice that the minimum needs to be in place for all. Some tools need to be available. Some governance to minimize failing with security or customer data needs to be in place. But that can be just the bare minimum.
P.P.S. ...and then there are industries where legislation sets a higher minimum bar.