Skip to content
The largest shift since the internet

Artificial intelligence

The goal is not simply to use AI. It is to understand where it genuinely creates value and where it carries risk.

I consider artificial intelligence the largest technological shift since the internet. I brought AI assistants into my daily practice early, not out of curiosity, but because the gain was immediate and measurable.

That daily use changes the kind of advice I can give. I do not discuss AI as a topic of analysis: I discuss it as a tool I use to produce real work, with specific strengths and specific limits.

The question is no longer whether to govern AI, but how fast — because it has already entered your organisation through the side door.

Where I help

Framing the use cases

Separating measurable gain from fashion. Most AI projects fail at this step, never at the technical one.

AI governance

Which data may leave, to which vendor, under which contractual terms, and who decides when a case is not covered.

Data sovereignty

Which jurisdiction processes your data, whether it feeds training, and what happens if the vendor changes its terms.

Agents and automation

Building flows that save real hours, with least-privilege permissions and a human on the irreversible decisions.

AI system security

Prompt injection, context leakage, abuse of tools exposed to an agent. An AI integration gets tested like anything else.

Responsible use

What can be delegated to a generative model, what must be reviewed, and what is never handed over. The line gets drawn before the incident, not after.

Tools and models

Used daily, not merely evaluated.

Assistants

  • ChatGPT
  • Claude
  • Microsoft Copilot
  • Sakana AI

Concepts

  • AI agents
  • Large language models
  • Assisted automation
  • Context engineering

Guardrails

  • AI governance
  • Data sovereignty
  • Acceptable use policy
  • Law 25

A point I often make to leaders: AI is already in your organisation. It arrived through personal accounts, browser extensions and features enabled by default in software you already pay for.

The first AI policy written in a hurry is almost always a ban. It achieves two things: it changes nothing about actual usage, and it makes that usage invisible. Better to start by looking.

A question on this?

Describe the context in a few lines. I answer with a proposed angle, not a brochure.