Knowledge
AI for mid-sized companies, explained
Answers to the questions that come up before deciding on a model of your own. Every article names its sources.
When AI text becomes visible: what companies now need to settle
Since 2 August 2026 Claude marks every text invisibly. What a hit proves, what it does not, and what companies actually need to settle.
As of 12 August 2026
Search documents or fine-tune a model: what each is good for
When does a fine-tuned language model help, when does document retrieval? Differences, limits and a decision aid for management and IT.
As of 10 September 2026
What an AI model of your own really costs
A three-year calculation: model, hardware, electricity and running effort, broken down per workplace. All amounts net, every figure sourced.
As of 14 September 2026
The AI Act: what applies to ordinary companies and what does not
Which duties the AI Act places on a mid-sized company, which deadlines apply after the Digital Omnibus, and what does not apply at all.
As of 12 August 2026
What hardware an AI model needs on your premises
How much graphics memory a 9 or 27 billion parameter model needs, which machines deliver it, what they cost and how to spot a suitable system.
As of 12 August 2026
Where your documents go while a model is being built
The path of a document from upload to deletion, who is involved in which role, which deadlines apply and which residual risk remains.
As of 12 August 2026
What a model learns from your documents, and what it does not
What a model fine-tuned on company documents picks up, where its limits are, and which measures against them are demonstrably effective.
As of 12 August 2026