Perspective, not noise.

For the past few years, leaders have been told to learn how to prompt AI. The harder question is now arriving: what work are you actually willing to hand over to it? Imagine an AI agent opening your CRM, reading customer history, preparing a proposal, updating the pipeline and scheduling the next action. Nobody is asking it a question anymore. Work has been delegated.

The most robotic conversations I had last week were with humans. And it made me question something we rarely ask in AI: Are we sure we should make AI as human as possible? Last week, on my road trip home, my car broke down in France. I was travelling with my three daughters. Late at night, after being towed away, we were dropped in Massiac, a tiny mountain village. No car. Luggage in hand. No idea where we would sleep. ANWB had the full case. They knew where we were and who I was travelling with.

متى يشعر الإنسان بأنه ينتمي حقا إلى وطنه؟ مداخلة قدمتها في ندوة بالرباط حول مواطنة مغاربة العالم في العصر الرقمي — وكيف أن الانتماء الحقيقي لا يقاس بالمسافة، بل بالأثر.

Your junior employees were the training ground. AI just removed it. For years the messy first draft, the clumsy customer reply, the analysis that took too long, all of it looked inefficient. But that was where judgment got built. Through small mistakes, unclear tasks, difficult customers, and watching experienced people make trade-offs in real time. Now AI does that first layer. The draft is generated, the meeting summarized, the research compressed, the reply suggested.

Forty percent of AI agent projects will be cancelled before 2027. The reason will not be the models. Gartner put a number on something I keep seeing up close. The technology works. The agents are capable. And still the projects quietly die. What kills them is rarely the AI. It is an agent dropped onto a process nobody fully understood. A workflow with no clear owner. A pilot that impressed everyone in the demo, then met the messy reality of how the business actually runs.

Everyone is using AI. That is becoming the problem. Not because AI is bad. Not because people should stop experimenting. The problem is that AI is entering companies faster than teams can structure its use. One person uses ChatGPT for strategy. Another uses Claude for documents. A developer uses Copilot. A recruiter uses an AI sourcing tool. A manager uses AI to summarise meetings. A sales team uses AI to write outreach. Individually, all of this can be useful.

AI agents are entering the messy part of business. Not the clean demo. The real business. The inbox full of half answers. The follow-up nobody sent. The candidate who replies late. The customer who changes the context. The manager who wants speed, but also control. That is where AI agents start to matter. Not when they give a polished answer. When they help carry the small operational loops that slow companies down every day.

Agentic AI is moving from assistant logic to execution logic. That may sound like a small shift. It is not. It changes how companies design workflows, governance, procurement and responsibility. For a long time, AI in companies was mostly about support. Support for writing. Support for search. Support for analysis. Support for faster thinking. That phase is not over, but something more serious is emerging now. Agentic AI is moving from assistant logic to execution logic.