The Full Circle: 20 Years Ago I Studied AI. Then I Left. Here's What Happened When I Went Back to School.

Photo de Tim Mossholder sur Unsplash

In 2005, as a young researcher, I was publishing papers on machine learning. In 2026, I'm sitting in a classroom learning to build AI agents. The gap between those two sentences contains my entire career.

The Departure

Twenty years ago, I started a PhD in artificial intelligence at Université Paris-Sud, working with CNRS and INRIA. Recurrent neural networks, dynamic multi-armed bandits, stochastic optimization. The kind of AI that lived in labs and academic conferences, not in products. I had the privilege to teach at École Polytechnique, Polytech Paris-Sud and EFREI. I loved it, until I realised research and engineering needed a bridge. Then I left.

In 2010, enterprise AI maturity was a marketing fantasy. The gap between what I researched and what companies could actually use felt unbridgeable. So I followed my curiosity toward something valuable: building engineering skills, projects and teams.

Over the next fifteen years, I went from writing algorithms to leading people. Software engineer, team leader, Scrum Master, various flavors of engineering management. I shipped products, I built organizations.

The AI researcher became an engineering leader. That wasn't a detour. It was an apprenticeship, and I had multiple opportunities to promote and support AI and data initiatives wherever I went.

The Quiet Thread

I never fully let go. Since 2011, I have been following closely the changes, the innovations, the transformations and opportunities around the emergence of Data Science and AI.

When generative AI exploded, I wasn't surprised, although change happened real fast once technology matured. I was ready. At my current company, we started exploring quickly: prompt parties, integration of OpenAI models internally, a secure coding assistant for software engineers, agentic AI experiments and more. I contributed my leadership to support innovation initiatives that connected what I'd researched decades ago with what teams needed today. The circle was starting to close. A new loop was opening.

But exploring isn't building.

The Return

Last month, I enrolled in Join Lion's "Build Your AI Agent" training. A PhD in AI, going back to school. Some people found it funny. I found it necessary.

What I expected: a refresher on agentic architectures, some hands-on labs, maybe a framework or a use case I hadn't tested yet.

What I got was different.

The technical content was solid. I dug deeper into agentic systems, automation patterns, and use cases that I'd been circling around on my own. But the real value wasn't in the slides. It was in the room. Practitioners from different backgrounds (founders, product leaders, curious explorers) all converging on the same conviction: AI agents aren't a feature. They're a paradigm shift. And the people who build them now will shape what comes next.

Being a student again after fifteen years of leading was humbling in the best way. When you've spent years making decisions and coaching teams, sitting down to learn forces a different kind of attention. You listen differently. You question your assumptions. You remember why you started.

What I Found

Three things crystallized during that training.

Pragmatism beats theory. I spent years in a lab optimizing mathematical functions. Beautiful work, but disconnected from production. The agentic AI space today is the opposite: messy, practical, immediately useful. I've been exploring tools like Claude Code for months, testing use cases that range from content automation to engineering workflows going through proactive agents. The training confirmed that this instinct was right: the value is in building, not theorizing.

People accelerate everything. I met founders who are shipping AI products, engineers who are rethinking their entire stack, leaders who are betting their strategy on agentic systems. The conversations I had in three days would have taken months to find on my own. Some of these people will show up again in my professional life. I'm certain of it.

The moment is now. I've been exploring AI for a while: reading, experimenting, contributing to enterprise initiatives. But exploring has a shelf life. At some point, you either build or you watch others build. The training didn't teach me that. It confirmed what I already knew: it's time to stop exploring and start shipping.

The Full Circle

Twenty years ago, I trained neural networks in a research lab. Today, I'm building AI agents that solve real problems. The path between those two points isn't straight. It runs through engineering teams, cloud transformations, leadership programs, teaching classrooms, and personal projects that nobody asked for but I couldn't stop maintaining.

Every chapter contributed something. The researcher gave me the foundations. The craftsman gave me the discipline. The leader gave me the ability to think at scale. The teacher gave me the instinct to make complex things clear. And now, all of it converges.

I'm building. Not from zero, but from twenty years of accumulated perspective. The kind of perspective you only get by leaving a field long enough to grow, then coming back with different eyes.

If you're on a similar path (a technical background that evolved into leadership, a passion for AI that never quite went away, a conviction that the time to build is now) I'd love to connect. The next chapter is more fun with company.

This post is the first in what I expect will become a series. I have deep-dives brewing on the agentic AI use cases I've been building and on what actually changed in AI over the past twenty years. Stay tuned.

Produced with La Redaction: Margaux (🧭 Strategy), Lucien (✍️ Draft & Polish), Camille (🔍 Review), Solene (👑 Approval)

Status: BON À TIRER — approved for publication