Why Generative AI Became Mainstream So Quickly
ChatGPT reached 100 million users two months after its launch, a record at the time. For Mustapha Benkalfate, the reason is accessibility: for the first time, anyone can use AI without a data scientist.
"It's as if we gave everyone the ability to fly a plane."
He immediately warned against the narrative from software providers, who promise a world where AI is everywhere and "frictionless." "These are fairy tales, you could also call them lies. Don't be too afraid for your jobs, as long as you master AI and understand what the tools can do."
The Four Use Case Areas
1. AI for All
Writing better, reading documents you never have time for, and drafting memos and summaries. His favorite tool for this area is the free NotebookLM, which turns PDFs, videos, websites, and transcripts into reports, infographics, or podcasts. He emphasizes its use for onboarding new employees and products:
"The first time something is explained to you, pressure and adrenaline prevent you from truly listening. The second time, you don't want to seem disengaged. An AI that answers at any time, without judgment or stress, helps you learn faster."
2. Data
Querying your data in natural language, without waiting for an analyst. Generative AI is not meant to replace Excel or Power BI for calculations; it helps you use them. To determine if a report generated by AI is trustworthy, he suggests a simple test:
"Open three different conversations and ask the exact same prompt. If the reports diverge significantly, you cannot trust them. If they converge, you can build upon them."
He also recommends always asking the AI to explain its step-by-step reasoning and data sources.
3. Specialized Tech
Debugging, ticket sorting, technical documentation, and image generation for marketing. He adds a point that concerns all companies: after SEO comes GEO (Generative Engine Optimization). "You must ensure that when an AI searches for information on the web, it finds your company and your products."
4. Training and Development
Everyone now has a private tutor. He gives his own example: having had almost no English classes in Algeria, he uses AI to analyze transcripts of his keynotes and meetings, group his English mistakes, and then turn them into a 100-question quiz. He calls this logic the "circular economy of knowledge": no longer letting what you say and learn go to waste.
Two Methods Demonstrated Live
SCOPE©. In one to two minutes, he built a prompt live on stage to prepare a keynote on generative AI risks for SME leaders in Hong Kong: specialist, context, objective, audience, execution. "It's not five seconds, it's more like a minute or a minute and a half. But that's what makes the result interesting."
SEAT©. He asked an AI to reverse-engineer the prompt that might have generated a photo of him. Claude refused, but Gemini produced the prompt. The description saw him as "in his late thirties": "I'm 43, that made my day." He then showed how to modify this prompt to produce a new image and explained why the method is in high demand in banking and insurance: starting with a successful document to extract its structure, then adapting it.
Shadow AI: Banning Doesn't Work
Answering a question on data protection, he said that two strategies fail: banning everything, and allowing everything without governance. When tools are blocked, about 80% of employees report using shadow AI on their personal devices. The only method that works, he argues, is education, supported by a network of ambassadors identified through dynamic assessment, representing about 4% of the workforce. Read our analysis of shadow AI.
AI Agents: Choosing Your Risk Zones
An agent is still a generative AI: probabilistic, prone to hallucinations, and a "black box" whose step-by-step reasoning no one can explain. He revisited a concept from his book, "profit maximization zones": you place agents where the stakes are real but the error is acceptable, such as certain sales processes, but never where an error is irreversible.
What Overcomes Resistance
People are reluctant for two reasons: fear of not being "tech-savvy" enough, and fear of being replaced. Both disappear, he claims, as soon as they are trained and understand the limits of the black box. He proposed an exercise for the leaders present during their lunch break: ask any AI how many hours of AI training a leader should complete each month, then compare it with their own schedule.
"Leaders don't want to replace everyone. They want to replace people who don't master AI with people who do. This isn't my opinion; it's what they tell me. I explain to them that you must transform people, not replace them, or you lose internal knowledge."
His conclusion: "Humans and technology together is success. Technology without humans is failure."