From AI Tools to Agentic AI

Why most organizations are optimizing the wrong things

I’ve come to 2 slightly uncomfortable conclusions after working on AI across Europe and more recently spending time and work across the Americas, speaking and listening (a lot):

1. Most organizations don’t have an AI problem. They have an ownership problem.

2. Many still treat AI as a feature.

But of course AI is not a feature. So what is it then?

That question came up in almost every meaningful conversation I had last month during my work outside Europe.

The uncomfortable truth inside organizations

Another pattern I saw repeatedly across Europe and North America;

IT thinks: “Business is not leaning in.”

Meanwhile, most organizations are still:

• improving processes

• automating tasks

• running “data + AI” transformations from the tech side

• Adding copilot. Rolling out ChatGPT licenses...

The real bottleneck isn’t where you think

Across markets, one thing became very clear: it’s not data; it’s not AI models; it’s not even integration. It’s decision friction across the lifecycle.

In asset-based finance (leasing and factoring), where I spend several years before AI emerged as the third tech wave, organizations are built around processes like:

• credit assessment

• invoice verification

• pricing

• collections

These are handled as separate workflows. But they’re not. They are connected decision moments.

And the moment you start looking at them that way, something shifts. This is where agentic flows begin to outperform traditional systems.

Most AI efforts today still focus on:

• improving steps

• adding tools

• supporting workflows

But real impact comes from something else:

Replacing decision-heavy workflows altogether

That’s where (1) speed increases, (2) risk becomes dynamic and (3) margin is actively managed

Not in dashboards but in decisions. And that’s exactly why most AI pilots stall. They optimize steps… instead of redesigning the decision flow.

A moment in Brazil

Last week, I was on stage in Brazil, speaking about AI. But the moment that stayed with me happened somewhere else. During a small, off-conference session, I worked with a mid-sized leasing company from Campinas. Not Big Tech and not another innovation lab.

Just a business running on deals, risk, and margin. Very recognizable also from an European perspective. Meaning, everything works but everything takes time.

We didn’t start by redesigning processes. We started with one simple question:

“Which decisions do we want AI to take over?”

That question changed the entire conversation. Within hours, we moved:

• From reviewing in hours/days → to better action in seconds/minutes

• From workflows → to decision flows

• from copilots → to autonomous agents

Especially the first change was an eye opener for many around the table:

It showed that 'Today' decisions are delayed by dashboards. Data was collected, analyzed, discussed and only then acted upon. Each step adds time, inconsistency and manual effort.

With agentic AI, that cycle collapses. The system interprets, decides and acts in one flow.

The results were not theoretical: (1) documents automatically interpreted. (2) risks instantly scored (3) deviations visible in real time (4) pricing supported by context, not just rules

Was it perfect? No. But it was fast, moving from days to minutes and more importantly:

It was fundamentally different. It was not optimization. it created (= redesigned) a different operating model.

This is not an IT project

There’s a lot of noise around “agentic workflows” right now. But in practice, the shift is simple:

It moves AI into the business, where value actually sits. Not in another AI lab. Not in a tech-only initiative. In this respect, SME have an advantage as an enterprise-sized mammoth.

What works are better are the short cycles where: business, data and AI models come together around real decisions, not use cases. That’s exactly what we structure in a Decision Sprint.

Why Brazil feels different

Brazil was interesting. Not because they are “ahead.” But because they are less constrained.

In Europe, we improve these processes by 10–15%. In Brazil, I saw companies replacing them.

Rebuild Decision Systems with one different question

But AI is not something you add. It’s something that replaces workflows.

Especially in areas like:

• credit decisioning

• underwriting

• onboarding

These are not “process improvements.” They are decision systems waiting to be rebuilt.

Which leads to a different question:

Not: Where can we use AI?

But: Which workflows should no longer exist in their current form?

.

What AI-native actually looks like

Very few organizations are structured for this but I saw one example that stood out.

Nubank.

They made a decision most European banks wouldn’t dare to make:

Becoming AI-native is more important than expanding globally.

Their ambition is clear: To become fully AI-native.

That means decisions sit in squads, not in management layers (risk, data, and product operate as one team), AI is embedded directly into workflows, not added on top!

What makes Nubank different isn’t the technology. It’s how they are organized: Small, autonomous squads, built around real problems.

And most importantly: decision ownership sits inside the team

AI doesn’t scale there through tools. It scales through how decisions are organized.

Even the US is still catching up

Interestingly, in the US, companies like NerdWallet and DoorDash told me they still need 8–12 months to fully embed AI into business processes. So this shift is not “solved” anywhere yet.

Why AI remains stuck in pilots

A quick check on your organization is:

• is it still siloed

• project-driven but not value-driven

• dependent on central teams

Then, AI will remain a pilot.

Not because the technology doesn’t work but because the organization isn’t designed for it.

And that brings me back to Europe where we still largely treat AI as a tooling discussion.

Where teams are structured around data + AI ( = tech-first). While the real shift is happening elsewhere.

Where this is going. The Next Thing.

I’m not fully done processing everything I’ve seen. It feels I am ahead as adoption is still around AI tools. Which is not wrong but pushing AI in business processes is the next thing but should be on a 8-12 month horizon like I hear whispering in the US.

One thing keeps coming back: Some companies (especially in Brazil) understand AI as a different game. They are not improving workflows. They are rebuilding them as real-time, AI-driven decision systems and if this is where it’s going…We’re optimizing the wrong things.

Most important insight I learned;

AI adoption = organizational design first. Technology second.

And that’s a very different conversation than most organizations are having today. Challeging but exciting time ahead in change management and digital transformation!