There are two classic ways you get stuck with AI and agents. As with any new and novel effort there are countless ways to fail. There are ways to succeed too. We start this time from the ways which are doomed to have only minor progress.

The intuition to roadmap and strategize is so inbuilt into our brains and operations that I see more than half of organisations I've had the privilege to watch fall into. The actions feel right. Both create papers and artefacts that appear right.

A: The initiative list

How this works: you ask around and list down all good ideas on what we could do with AI. Have interviews and maybe hold few workshops. Write physical or virtual post-its. Group and plot. Probably do impact/effort. You will soon end up with 10-50 ideas or initiatives. Then you have your leadership look at the results. Good ideas.

The list is there - but what then? I find it both curious and logical at the same time that no progress is being made. The ideas don't cross into hearty execution. The ideas are there but intrinsic and big-time pull is not happening.

But I suspect you might already see the disconnect.

B: Strategy without soul

Another failure mode is to go think-through mode and start strategising. Get managers into the room. Take on an established analysis framework. Probably not SWOT but whichever is your favourite.

Discuss the situation. Map out the current state and score readiness.

You arrive at a situation where the analysis does not feel right. The feeling is that of skating on thin ice. Hmm. Maybe that is not the right feeling. The feeling is that of trying to project the future that feels like un-guessable. You don't know what to latch on to. You feel there are too many unknowns. The more you think about it, the more there is uncertainty of what really works.

The act of classic strategising cannot remove the core uncertainties of the fast-moving future. You cannot even really pin down what is relevant.

It feels like catching something that does not want to be caught. You might notice why the soul is missing.

Both failing similarly

Neither of the starting points builds the one thing that matters: the organization's ability to see what's possible. You simply cannot see the right next target state. However much you squeeze your thinking muscle it still does not feel good strategies come out. You might have flashes of something. But you cannot trust what you see is the right thing to execute. You can't prioritize what you can't imagine. You can't govern what you don't understand.

Both approaches fail because they try to think their way to answers that only come from doing.

Then you might realise this is both a complex future as well as fast-moving domain. The right action here is an action of deliberate learning.

The right sequence

The agentic thinking is primed with 3 elements according to my experience, but also backed up by leading practitioners out there. Skills + agentic tool + WoW-moment = path to exponential.

People like to lay out a plan as sequence. In real life, events intermix and this kind of clear-cut sequence does not really happen. But we can learn a lot and we can use this to guide ourselves on the high level.

Skills unblock

The LLM is not an automatic miracle-maker. How you use the LLM directly affects what you get. If you are satisfied with chatting, keep on chatting. If you want something else like agentic behaviour, then you need to ask for it the right way.

The act of turning an LLM into something agentic is a skill. Knowing what is agentic is a skill.

Getting 20 people building real agents on your own real problems is an actionable first step. This in turn leads people who've built an agent see their own processes differently.

The truly agentic tool

You can twist and turn and make the vanilla ChatGPT, Claude or Copilot chat into an agent. But it will still feel like a chat. People are very used to the chat-paradigm so doing more that will not make for profound change. You need something else.

I have been watching the progression of the code-generating agents (Claude Code, ChatGPT Codex) make headway for the last 9 months. Ever since original release at spring of 2025 these agentic tools have been reshaping how we approach problem-solving. While people think these tools are for code-generation, these generalist agents are actually the clearest path to general agentic future.

The code-generating agents (Claude Code and Codex) are the tool of today that create the WoW-moment. "Shit. I can really do that." These moments virtually all the time are with slicing and dicing enough internal data (PowerPoints, excels, database content, ...) and making something that was weeks of work before. All of a sudden, you have results at hand that appear correct.

WoW-moment

So: you basically bring people the WoW-moment as soon as possible. This moment reframes how people see the possibilities now and in the nearest future. The WoW-moment changes a person from passive believer to active doer.

The WOW moment isn't educational — it's strategic in the way that it changes the pace of learning. Chatting and following the media does not teach you enough of what is really possible. Just like watching football does not teach you to play football. To learn to play, you have to play.

The priming ingredients are the skill, the agentic tool and the wow. Thereafter everything else can be just work. Not easy, but not mystery either.

Discovery follows

Your people come back with the initiatives that should have been on the list. Better yet, based on my experience, everyone starts building eagerly and without asking. The innate desire to build has always been there.

Now you have 20 people who can see. They look at their 200 processes and know which 5 to try.

The first discoveries are probably wrong again as usual. You cannot escape the failing first. The jagged frontier will catch people. The brain learns from repetition. The brain learns from both success and failure.

But you have your people on at best as daily learning cycle. That is quite different than listing initiatives on monthly basis!

The right discoveries are embedded in the deep context of your organisation. You need discovery at scale to work though the bigger change.

Context gets evident

You notice that you cannot easily get your organisation knowledge available for your agents. You notice you need to fix that at scale. You notice the data access operational data is not there yet.

You notice you just need to build the verification infrastructure. The difference is that now you have examples of agent outputs and even decisions that need verification. This is no longer an abstract discussion. The same with agent governance. You notice you need to have a process for security, sharing, and for instance data handling. This is now easier as you can respond to gaps, problems and worries grounded in real cases.

What was seen as a generic enabler is now seen not as a prerequisite — but as a response to what discovery revealed.

Platforms become defined

People often say that we can't have agents without a platform. So, you end up chasing your tail choosing the best platform for months. But you also realise the landscape moves on weekly basis.

Again, this choice has been stuck until you've gotten moving. Once you are moving, then this question becomes practical.

The platform question was fully unbounded before you got going. Now you can frame the question as: "What do I need to get to the next level?".

Momentum creates clarity.

The right sequence feels risky (but isn't)

The conventional wisdom says: strategy first, then execute. I have executed this before. There will be comments. "We should be more mindful about this". Or "I don't know where this is going". All of this feels risky because you're "spending time on training" before you have a strategy. You are also building what you don't feel like you understand.

People want to know where all of this is going. I like the term AI literacy. It nicely captures the ability of a person or organisation to verbalise what AI can do. The strategy can't come first because the strategists don't have the neural pathways yet to think about the future with AI.

People still claim the risk is there that we just are headed the wrong direction. I really like to flip the model. The question is: "What do you think the future is?". No answer required. Just personal reflection on where this explosive growth of AI-capability and crunching power is going. The large majority knows now that both opportunity and risk of staying still are there. The personal will to act is there.

What is the frame that gets us moving fast?

The real risk is the other way: 6 months of roadmapping, then half-hearted progress here and there. But mostly just coming back to list the good ideas again.

The companies that move fastest aren't the ones with the best strategy deck. They're the ones where 50 people can see what an agent can do — and the right initiatives emerge from the organization, not from a steering committee. You're not delaying strategy — you're making strategy possible.

My own take is a broad movement. By broad I mean literally everyone in the company. But you don't build a real movement by starting with everyone on day 1. Think how people actually want to join. The skill is both the catalyst and the reward.

You can build a strategy around that.