Hello!

The last time I wrote was 2019. This is ages ago! So many have many things have changed. Most notably, in our line of work, the advent of AI. Or more specifically, the rise of the Large Language Models.

And yes, I am a power user. I use Claude Max, Claude Code, Github Copilot Enterprise and occasionaly Perplexity and ChatGPT. Lately I also dabble with Microsoft Scout and M365 Copilot.

I have created my own agents and skills and can now delegate a great part of my work (consultancy, coding and platform engineering) to a model.

The dark side of commercial LLMs

However, I see the dark side too. My work is done faster, but my brains are getting tired and less happy. So that is why I started to write this blog post myself. No AI here, pure me.

Also, I see the dangers of (over)sharing with large language models. We share way more with commercial LLM's, than we do via Social Media platforms or via Google in the olden days.

You submit prompts in plain text, and while commercial AI providers often claim they do not use your data for training, they still collect your prompts, documents, and images. This means the complete context of your input is stored and processed on their systems. It is not always clear how this information is used, but it is reasonable to assume that such data holds significant value for them.

That is why I invested in a MacBook M4 Pro Max with 64 GB of RAM, to run these models locally, even in an air‑gapped setup. Personally, I believe this is something more people should consider.

Flow chart: to prompt or not to prompt

This flowchart helps you decide whether it is safe to share a prompt, document, or image with an external AI service. If the content is sensitive, keep it local or air-gapped. If it is not, share it only after checking the provider’s privacy and data-handling policies

Next time will be a more technical post about which local model to choose, how to run and what to tweak.

See you then!