Try it.
Ideas become interesting when they meet reality. We build, test and pay attention.
AI Lab
Experiments first. Business model later.
The best way to understand new technology is to work with it.
We test, combine, measure and discard. An experiment may become a product or a client project. It may just as well show that an idea does not work.
That is a result too.
A real-world lab for artificial intelligence and natural common sense
AI on.
Brain engaged.
We test what AI can do. And take responsibility for what we make of it. With curiosity, room for disagreement and a willingness to discard our own ideas.
Not immune to stupidity. Just remarkably good at surviving it.
Question it.
A convincing answer is not yet a reliable result. We test assumptions, look for counterexamples and make limitations visible.
Own it.
AI can suggest and do the work. Direction, judgment and responsibility for decisions stay with us.
Why an AI Lab?
Generative AI is developing too quickly to judge its possibilities from product announcements, benchmarks or demonstrations alone.
So we work with the systems ourselves.
We build real workflows, test them under practical conditions, and observe where they work reliably — and where they do not.
We are interested in more than the visible output.
Just as important are:
- repeatability
- control
- quality assurance
- orchestration
- data and measurement
- cost and production effort
- human decision-making and responsibility
From AI tool to AI-native system
A single AI model rarely solves the whole task.
That is why we combine language, image, video and voice systems with conventional software, local models, cloud services, data and specialized AI roles.
More AI does not automatically mean better.
More AI ≠ better.
Deterministic tasks can often be handled more reliably with conventional software.
AI becomes especially interesting where context, interpretation, exploration or creative decisions are required.
The actual experiment therefore often begins between the systems.
Specialization instead of a universal bot
We increasingly test AI-native production systems in which different roles handle different tasks.
One AI may research and develop counter-models.
Another can orchestrate visual systems and check results.
Other systems handle data, measurement or recurring operational processes.
Humans set direction, taste, boundaries and final decisions.
Not one AI. A system.
AI as an ideation partner
One question interests us in particular:
Can AI do more than produce existing ideas faster — can it open new search spaces?
Twenty variations of the same idea are not creative exploration.
It becomes more interesting when a system:
- proposes new axes
- changes rules
- finds unexpected combinations
- develops counter-models
- or opens a solution space that has not been considered before
Selection remains human.
But the idea no longer has to come exclusively from a human.
Learning under real-world conditions
The test does not end with one successful image or video.
A system becomes interesting when it works repeatedly under changing conditions.
That is why we consistently distinguish between:
Observation → Interpretation → Decision
Not every correlation is a cause.
Not every plausible explanation is already a fact.
And not every failure needs a new theory immediately.
More experiments
More experiments will appear here when there is something genuinely worth showing.