Skip to main content

36 Years at IBM, a Centralien Engineer, and the AI of Tomorrow

A conversation with Olivier Casile – IBM Distinguished Engineer (Retired) • Former CTO of the IBM Global Industry Solution Center, and IBM Client Technical Leader for a major international insurance company.

By Louise Servoin · 2026-05-05 · 10 min read

## Olivier Casile – IBM Distinguished Engineer (Retired)

Former CTO of the IBM Global Industry Solution Center and IBM Client Technical Leader for a major international insurance company. Graduate of École Centrale de Nantes, PhD in Telecommunications, expert in AI, Enterprise Architecture, Cloud Architecture, and Quantum Computing. Based in the Nice region, France.

Disclaimer: The views and opinions expressed in this interview and associated posts are those of the author and do not necessarily reflect the official policy, position, or views of IBM.

Some careers resemble sprints. Others are like long journeys through successive revolutions. Olivier Casile belongs to the second category. Joining IBM in an era when code was still written on green terminals, he left 36 years later—after leading and participating to a number of IT revolutions: e-business & business integration, Hybrid Cloud, Enterprise AI, and Quantum Computing.

Now retired, he maintains a sharp perspective on what companies do (or don't do) with their accumulated knowledge, and on what AI can change… if we truly put in the effort and manage the risks.

## A Career of Technological Revolutions

Olivier, you spent 36 years at IBM, from the mainframe era to GenAI. How did you experience these technological transitions from the inside?

Every revolution had its promise and its resistances. What struck me was that the real difficulty was never technical. Engineers innovate, solve problems, and adapt quickly. The true challenge was always human: how to convince an organization that its skill base had to evolve, that its processes needed to change, and fast?

"The real difficulty was never technical. Engineers adapt quickly. The real challenge was always human." — Olivier Casile

You were CTO of IBM’s Global Industry Solution Center in Nice-Paris, and Global Technical Director for a major insurance group. What did this teach you about how large organizations absorb (or fail to absorb) knowledge?

In my role as an IBM Distinguished Engineer, I had a unique vantage point: I saw both IBM Research labs (places of pure innovation with Nobel Prize-winning researchers) and I was leading a field team who had to integrate these innovations into the IT landscape of large international companies. The gap between the two was often staggering. Not for lack of need or will, but for lack of a structured bridge between the experts and the practitioners.

## AI as an Amplifier — or a Mirror

You closely followed the emergence of Watson (™), then LLMs, and now GenAI. Where do we really stand in AI's ability to transfer expert knowledge?

Watson (™) was an extraordinary promise, and the story continues at IBM. It was, and still is, a vast set of innovative technologies, often ahead of its time. I saw it in action in complex use cases in banking, insurance and various industries. But we often tried to skip steps: deploying the technology before structuring the knowledge we wanted to entrust to it, or planning the changes and impacts on organizations and skills.

What I observe with current LLMs is a similar risk: the temptation to apply excessive and costly solutions to every use case, because the fluidity of the language they produce creates an illusion of competence. An AI can appear expert without actually being so on critical cases, for instance by wrongly claiming it knows an answer, or fabricating elements of proofs, a phenomenon known as hallucinations.

"The fluidity of language creates an illusion of competence. An AI can appear expert without truly being so in critical cases." — Olivier Casile

In your experience with large-scale technical knowledge transfer—having given over 25 major scientific presentations—what makes the difference between training that sticks and training that evaporates?

What sticks is what we practically used to solve real problems. When I explained data architecture and governance, machine learning, generative AI to enterprise leaders or medical physicists, theoretical presentations had limited impact. What worked was practical experience: presenting a critical case and asking the audience to reason out loud to defend their position. This is exactly what a good AI training system should do: not give theoretical answers, but force the construction of practical reasonings.

## Quantum Computing: The Next Skill Disruption

You have written on Quantum for your clients—use cases, error correction, programing frameworks, quantum-safe encryption. How do we train professionals for technologies whose full implications won't be visible for 5 or 10 years?

This is a true pedagogical challenge of our time. Quantum is not an update, it’s a paradigm shift, a different way of solving complex problems. When I explained to CISOs that their current encryption algorithms would be vulnerable in 10 years, the first reflex was to push the problem away.

Solving problems with Quantum technologies requires new levels of skills (Quantum physics, sophisticated mathematics) which take years to acquire, hence you need to build skills and prepare your organization now in order to be ready in 5-10 years.

Regarding data, there are finance and healthcare regulations requiring that customer and patient data are securely stored for tens of years; then, even though the Shor’s algorithm can’t actually break current encryption keys, it will happen and you need to identify your data at risk and start storing it safely right now. Basically, training for uncertainty requires a different method: even though there are no immediate answers, you need to anchor frameworks capable of welcoming disruption.

## What AI Can (Really) Change in Continuous Learning

If you had carte blanche to redesign the training of engineers and technical experts—using AI as a lever—what would it look like?

It would look like the best mentor I had at IBM Research. Someone who doesn’t give you today’s answer, but anticipates, asks the right question at the right time. I had the chance to visit IBM Research labs in Yorktown and Zurich—environments where intellectual stimulation is permanent. Unfortunately, those resources are scarce and quite busy; this type of interaction is impossible to scale by human means alone.

But an AI system anchored in a real expert corpus, which “understands” the underlying challenges of your industry and speaks its language—that is where the promise becomes reality. Current LLM’s, multi-mode models, reasoning engines, prompting best practices, are just an intermediate step. Generating text, images and videos won’t suffice for long, the AI community is already exploring World models which will have a deeper understanding of the world rules behind text, images and videos.

Another direction of progress are small and medium models focused on specific domains, that can address the specific problems of an industry or of an organization at a much lower cost and more sustainably than large general models.

"The best mentor doesn't give you the answer. He asks the right question at the right time. That is exactly what AI should do." — Olivier Casile

You followed IBM watsonx.governance (™), Granite (™) models, and the rise of GenAI in enterprise. Is there a risk in delegating the transmission of expert knowledge to generalist AI systems too early?

There is a potential risk, like for any disruptive technology. That risk must be mitigated with appropriate governance: the right approach is to anchor the AI in curated and validated expert content—guidelines, real cases, institutional knowledge—and to preserve the specificities that make human mentorship valuable.

IBM, for instance, applies a comprehensive governance with thorough tests to guarantee the high quality of the data used to train its Granite (™) models. The goal of AI is not to model and replace the experts, but to complement and scale them. The nuance is vital, literally, in the healthcare domain.

## The “Wisdom” Angle

After 36 years of building and transmitting cutting-edge technical knowledge, what would be your most counter-intuitive advice to a company that truly wants to train its teams with AI?

Over years I have observed a recurrent pattern: the belief that once you have the tool, the problem is solved. Tools are necessary, of course, but they are not the most critical thing you need to solve problems! You shouldn’t start with a tool before deeply understanding what you will apply that tool to.

Start by mapping what your best experts know that is written nowhere else. This tacit knowledge (the way a senior reasons through an ambiguous problem, the shortcuts they use, the traps they avoid) is where the real value lies. AI, which is a tool, can only deliver value if we give it good material. Without it, you are just automating mediocrity.

My experience is that viewing AI projects through a pure technical lens is a recipe for missing the objectives, for disappointment.

"WATSON", "IBM watsonx.governance" and "GRANITE” are registered trademarks of International Business Machines Corporation (IBM).

Tags: Continuous Learning, Enterprise AI, IBM, Quantum Computing, Knowledge Transfer, GenAI, IT Architecture, Technical Leadership