How Allison Went from Sociology to Head of AI After Being Recruited in a Bar

My name is Allison Bon, I’m 28 and based in Lyon, France. I head the Artificial Intelligence department at XEFI, a French company specialising in local IT services for small and medium-sized businesses.

My role revolves around three pillars: leading the AI development teams and managing projects through to client delivery, driving AI literacy and training across internal teams, and conducting internal audits to identify where AI integration can genuinely streamline workflows.

The goal of those audits is always the same: save teams time on low-value tasks so they can focus on the work that actually requires human intelligence.

It is a role that combines technical management, hands-on teaching, and process optimisation, and no two days look the same.

I started with three years of mathematics at university, but pure theory eventually lost its appeal. I felt a growing need to understand the human side of things, how people interact, how society functions.

That pull led me to sociology, where I completed my master’s with a thesis on a question that still fascinates me: given that video game rules are identical for every player, why do we observe such completely different behaviours and practices from one person to another?

My original plan was to continue into a PhD and become a professor-researcher. What happened instead was that I landed in AI, and looking back, the sociology background turned out to be the most useful preparation I could have had.

Understanding how humans adapt to new systems, how structures shift, how technology redefines our connections: that was already precisely what sociology trained me to analyse. AI is simply a new and extraordinarily dynamic lens for the same questions.

The story is almost unbelievable. I was recruited in a bar. I met a woman there and we started talking about our lives and careers. She mentioned she informally scouted talent for her company and that they might be looking for a project manager.

Since I love an adventure and am never afraid of the unknown, I agreed to an interview just to see. It was only once I got there that they told me the role was for an AI project manager.

My knowledge of the subject was very superficial at the time, so I was honest with the CEO about my doubts. He looked at me and said something I will never forget: you spent nine years in higher education at the ENS, you can’t tell me you don’t know how to learn.

I accepted the challenge, worked relentlessly every single day to master the field, and after starting as a project manager I am now Head of the AI department. The company took a chance on my potential, my logic, and my adaptability.

In a way, AI is what landed me the job, and once I was in it, I went into overdrive to rise to the challenge.

I dived deep into theory first. I filled notebook after notebook: every time I encountered a new concept or technical term, I would analyse it, write down its definition, and try to build logical bridges to what I already knew.

My sociology research background was genuinely useful here, especially for tackling complex AI research papers without feeling intimidated.

I was also fortunate to have my partner by my side, a developer and technology enthusiast who challenged me, guided me, and most importantly pushed me to experiment on my own.

If I could do it over, I would never separate theory from practice the way I did. I spent too much time accumulating theoretical knowledge before applying it.

The real breakthrough came when I started working on concrete, hands-on projects. If I had started coding and testing tools from day one alongside my reading, my learning curve would have been much faster.

Confusing truth with plausibility when it comes to large language models. People use them as search engines, when they are actually statistical models. AI does not care whether its answer is factually accurate: every generated word is simply the result of a statistical probability calculated from training data. A sentence can be perfectly fluid, logical, and convincing while being completely false.

We must approach these tools with immense critical awareness and stop projecting our definition of human intelligence onto artificial intelligence. The machine calculates. It does not understand.

My own early mistake came from exactly this misunderstanding. I used to spend hours chasing the magic prompt through endless back-and-forth corrections.

The turning point came when I stopped scratching the surface and dived into prompt engineering and the underlying logic of these models. An LLM needs context, constraints, roles, and structure to truly perform. If a generative AI seems useless to you, the AI might not be the problem.

The trap with AI is that it allows you to do almost anything, which means you can quickly get lost, scratch the surface of dozens of use cases, and end up feeling like you are doing everything and nothing at once. To generate revenue, you need a strict framework and one specific use case where you become an absolute expert.

I always recommend finding what I call the sweet spot: instead of chasing a revolutionary idea you saw on social media, cross-reference a sector or skill you already master with a repetitive, time-consuming, or complex task that AI can drastically accelerate.

The goal is not to become a generalist AI consultant, which means very little today, but to become the person who solves a specific problem for a well-defined audience. By applying your expertise and your prompts to that particular pain point, you are no longer selling a vague technology. You are selling a concrete, immediately profitable solution.

When you try to do everything, you end up doing nothing.

There are so many tasks ripe for automation that it becomes difficult to choose what we actually want to delegate, and this raises a profound sociological question. Certain simple tasks allow our brains to pause and breathe. If we eliminate every routine micro-task from our daily lives, we are left with nothing but high-cognitive-load missions. The risk is genuine cognitive overload.

In our ultra-competitive world, time freed up by AI is rarely converted into rest. The reflex is to immediately reinvest it into even more demanding work.

It is therefore crucial to target only what I call brain-draining tasks that offer zero added value: routine email management, syncing complex calendars to schedule a meeting, building tedious spreadsheet macros. AI can already handle all of this perfectly.

Use it to eliminate frustration, not to saturate your mind.

We are in the era of agentic AI. It is what the entire industry is focused on right now, and the greatest potential lies in creating and managing ecosystems of autonomous agents.

For women starting out, there is a real opening in roles like AI Orchestrator or Agent Producer: packaging turnkey solutions for businesses, building workflow automation services where multiple AI agents collaborate to handle customer support, automate accounting, or plan entire marketing campaigns.

This is no longer about knowing how to code. It is about knowing how to structure logical processes and translate business needs into clear instructions for these agents.

There is also significant potential in creating specialised digital products, such as agent kits configured for a specific professional niche, which can generate recurring revenue by selling pure efficiency to overwhelmed professionals.

The barrier to entry has dropped dramatically. What matters now is clarity of thinking, not technical credentials.

Without hesitation: accepting and internalising the fact that I am deeply legitimate in my role and deserve my seat at the table. It is a daily psychological challenge, especially when you are hyper-aware of the gaps in your initial technical background.

On a day-to-day basis, I collaborate with and manage engineers and experts who have been immersed in this field since the very beginning of their studies. Finding your voice, asserting your leadership, and understanding that your value does not lie in coding faster than them but in your strategic vision and your ability to connect technology to human needs: that is a significant personal triumph.

My message to any woman in a similar position is this: be exactly where they least expect you. Legitimacy in AI is earned solely through action. Publish on LinkedIn, say yes when you are invited to speak, claim the space you want. No one will give it to you willingly.

The technical barrier to entry has completely crumbled. Today, the most powerful programming language for using AI is simply natural language. AI does not expect you to know how to code. It expects you to know how to articulate a need, ask the right questions, and structure your thoughts.

These are qualities of communication, clarity, and empathy where backgrounds in human sciences, sociology, coaching, or any people-facing field naturally excel.

In this field, the finest posture to adopt is that of Socrates: all I know is that I know nothing. Nobody has the absolute truth here. The landscape reinvents itself every single day, and whether we are experts or beginners, we all still have so much to discover together.

The only difference between the one who succeeds and everyone else is simply the courage to start. Do not wait for anyone’s permission.