AI is a tool. We treat it like one.
In it since 2019 – and never once caught up in the hype.
An extremely powerful tool, without question. But neither the downfall nor the salvation of humanity. We use it where it delivers real value – and not where it doesn't.
Between Hype and Fear
Few topics are moving as fast right now as AI – and few are simultaneously so hyped and so feared. Some expect it to solve all problems, others predict the end of the world. In between stand many small businesses that simply can't keep up and are wondering what any of this is actually relevant to them.
We got into the topic in 2019 – years before the ChatGPT moment that turned AI into a household name overnight. That time shaped our perspective, and it has stayed sober: AI is an extremely powerful tool. Nothing more, nothing less. Like any good tool, it delivers its value when used purposefully and correctly – and achieves little when used just for the sake of using it.
Tool, Not an End in Itself
A circular saw is great for cutting a plank – and the wrong choice for driving in a nail. With AI, it's no different. The art lies not in slapping "AI" on everything, but in recognising where it delivers real value and where a simple, transparent piece of program code is the better and more stable answer.
We make exactly this distinction on every project, time and again – with common sense and years of hands-on experience. The goal is never "as much AI as possible", but a stable, effective balance that will still run reliably in two years' time.
"Something with AI" – but where exactly?
Most small businesses feel the same vague pressure: you should "do something with AI". But rarely does anyone tell you exactly where it pays off – and where it's just expensive gimmickry. Instead, there are grand words and tools looking for a solution rather than solving a problem.
We flip that around. We look at where time is actually being lost in your day-to-day work and tell you honestly: it's worth it here, not there. A few areas where it almost always pays off in practice:
Invoices that file themselves
Scan a receipt, done. The AI recognises the sender, amount and date, tags and sorts it automatically. No manual typing, no searching through folders.
The archive that answers
"When did the contract with supplier X expire?" – asked once instead of ten minutes of flipping through files. Your own knowledge, made searchable.
Long documents distilled
Contract, official letter, expert report: the essentials summarised in seconds, instead of reading everything yourself from start to finish.
Texts that don't start from scratch
Product description, service copy, a draft reply – a proposal to review rather than staring at a dreaded blank page.
A website everyone understands
Content translated cleanly into other languages – not clunky word-for-word, but in a way that actually reads naturally.
Enquiries that sort themselves
Incoming messages organised by topic and urgency, routine cases prepared in advance – so you can focus on what matters.
What we use AI for
- Routine work that eats up time: classifying documents, extracting data, summarising long texts, tagging, translating.
- First drafts: text suggestions, structures, code scaffolding – as a starting point that a human reviews and completes.
- Patterns in large volumes: where a human would need hours and the result still remains verifiable.
And what we deliberately don't use it for
- Code we don't understand: We don't ship anything whose inner workings we haven't fully grasped. We are not a "vibe-coding shop" that can no longer explain its own code.
- Sensitive data sent to arbitrary models: What is confidential stays local – on models running at your site or ours.
- Decisions without human review: AI proposes, humans decide. Responsibility cannot be delegated to a model.
The Problem First, Then the Tool
For us, the starting point is never the technology – it's your work. We don't ask "Where can we plug in AI?", but "Where does it hurt right now?". Sometimes the answer to the problem is an AI model. But often it's a simple automation, a better form, or simply a clearer process – with no AI involved at all.
This honesty occasionally costs us a project that could have been bigger. But it's the reason why the things we build actually work afterwards – and why our clients believe us when we say: "You don't need that."
We understand the full stack, not just the button
For us, AI is not some external service we merely "call". We operate our own AI infrastructure and work with local models that run directly on the client's premises – on energy-efficient hardware, without a single document ever leaving the building.
As a foundation, we use open models such as Mistral and Gemma, and where a general-purpose model is too coarse, we fine-tune them specifically for the task at hand. Training our own models from scratch is something we have done in the past – today, for the vast majority of applications, it is simply a waste of energy and money. A good open model, properly adapted, is almost always the more sensible path. Making exactly these kinds of trade-offs is our actual job. And where a task demands genuine reasoning, we deliberately bring in a more powerful model – on our own infrastructure or via a provider with a transparent data policy.
This depth is more demanding than a quick API call. But it is the difference between "we just use some ChatGPT" and "we can tell you exactly which data flows where and why". Anyone who wants to know what's happening under the hood gets an honest answer from us – no magic, just technology that can be explained.
AI in your business – assessed honestly.
Wondering whether and where AI would actually make a difference for you? We give you a sober assessment – no hype, no fear, and no selling you something you don't need.