.svg)
Everyone seems to be at the forefront of automation. Or does it just seem that way?
I'm opening LinkedIn. My RSS feed is overflowing. Professional news is piling up. The same feeling over and over again: we are behind.
Companies automate half organizations. Finance is being replaced by agents. Development teams seem redundant. Marketing is now one intern with an arsenal of SaaS tools.
To be honest: FOMO is looming.
And then we try the same thing.
The tools work, only up to 80%. The “hacks” are generic. The vibe-coded solutions from agentic code editors? They still need a senior developer to clean up bugs, side effects and structural errors.
So what's wrong here?
Aren't we smart enough? Not fast enough? Not fanatical enough?
A statement by Kerry Moran (VP Research, Nielsen Norman Group) stuck:
“Smaller, more narrowly defined AI features are much easier to use. They may not be flashy, but they often make a difference.”
We recognize that. We are now seeing it in dozens of organizations.
The Great Pacific Garbage Patch of AI Deployment
AI promises a lot. But there is a serious risk in overuse.
Vibe-coded software runs business-critical processes.
AI customer agents crash when a question is slightly more complex than a shipping status.
Marketing content is full of generic, factually inaccurate claims.
The consequences of this will remain visible for years. Like the Great Pacific Garbage Patch, our careless use of AI will haunt us for a long time to come.
How we learned to stop worrying
We use AI on a daily basis. It is indispensable. We recommend it for almost every customer project.
But we don't focus on quick tricks or cheap wins. We focus on places where we have control:
- Deep operational knowledge
- Control over input and output
- Always a person in the loop
For each AI implementation, we go through the same checklist.
1. Can we define the task clearly?
Can we explain this step by step to a junior colleague?
Are the results measurable? Can we set quality criteria?
Would we dare to send the result directly to a customer?
If the answer is no, we won't automate it (yet).
2. Do we understand the unique cases?
Automation becomes exponentially more difficult when every second run results in an unexpected scenario. This not only leads to errors, but also to additional manual repairs.
We first identify edge cases.
If there are too many unknowns, we'll pause.
3. How do we keep people running?
AI should make people work faster and better, not invisibly replace them.
As soon as you do the latter, you lose control over critical processes.
We design explicit control points: reviews, approvals, decision moments.
People remain ultimately responsible.
What this looks like in practice
Every day, a customer received incoming letters that needed to be processed.
Each letter had to be read, the addressee determined and the correct account linked. Five manual steps per letter. Slowly. Error-prone at peak times.
We built a prototype:
- AI reads the letter.
- Determines the addressee.
- Links the correct account.
But: the employee always checks the final assignment before processing.
The process went from five actions to two.
The workload fell drastically.
Not an autonomous agent. No magic.
A narrowly defined tool, with a person in the loop.
Random cases are intercepted. The organization remains in control.
Not a revolution in weeks, but improvement in steps
We don't expect a digital revolution within a month.
We apply AI incrementally: testing, adjusting, improving.
This is a Kaizen approach.
AI is deeply embedded in processes, not superficially “sprinkled over” them. As a result, solutions become sustainable.
The ideal future: AI that actually 'works'
When AI is deployed step by step, we always see the same pattern:
Faster ROI
Value becomes visible in weeks, not quarters. Adjustments are made early, not after six months of failure.
Less technical debt
No vibe-coded solutions that keep haunting your codebase. Not half a job that needs continuous supervision.
Support in teams
If AI alleviates work instead of threatening jobs, adoption comes naturally.
Measurable impact
No vanity metrics, but a demonstrable reduction in workload and lead time.
The core
The AI revolution isn't about replacing your team overnight.
It's about finding small, solvable problems. They quantify. Understanding edge cases. And keep people running.
You are not behind.
You are careful.
And in a year, when a large number of companies with vibe-coded solutions are stuck in technical debt, you'll have systems that stand up.
Not sure if your next AI project has a chance of success?
Go through the three questions above with your team.
If you can answer them clearly, you are ready for a prototype.
If you can't do that, you've saved yourself six months of cleaning up.
