Evaluating the opportunities and risks for AI adoption: “We’re building trust with data, AI models, and governance” – Vasques

Globally, artificial intelligence has become a formidable force, transforming major industries and organizations, while also posing existential risks in several areas such as cybersecurity threats, data privacy issues, intellectual property infringement, among others…

In this interview, Xavier Vasques, Vice President, Chief Technology Officer, and Director of R&D at IBM France, talks to Ndubuisi Micheal Obineme, Managing Editor, The Energy Republic, on the role of artificial intelligence for the future of work, and how IBM is using AI to transform operations, build ‘Trust’ within its client-base in France and globally. Excerpts:

TER: What emerging technologies, such as AI, digital twins, automation, and advanced analytics, would have the greatest impact on the future of work?

Vasques: All of them will play a role, but the biggest shift is agentic AI. It is moving from assistants that answer questions to AI agents that take action across systems, under human supervision.

The real impact comes less from any single technology than from combining them. This includes foundation models for reasoning, automation for execution, and digital twins and analytics to ground decisions in real operational data. The challenge enterprises face isn’t just building one good agent; it is running a growing fleet of agents safely.

I’m not worried about AI adoption because AI raises our productivity and takes over repetitive tasks, allowing people to focus on what really matters. An engineer can spend more time on engineering, while the HR professional will focus on developing people rather than paperwork, and the finance team focuses on analysis rather than data entry. This is augmentation, not replacement. AI makes things easier for people to use their judgment, creativity, and expertise. With this, the future of work becomes a human setting the intent while agents handle the execution.

TER: How is IBM building trust with its AI solutions?
Vasques: This is a very important topic for us, and trust comes from three things: openness, governance, and control over your data.

On openness, our Granite models are released under permissive open-source licensing, with disclosed data practices. This matters because most large AI models are trained by scraping enormous volumes of data from across the internet with little control over their origin, which raises real intellectual-property and copyright questions.

At IBM, we take the opposite approach: we know what goes into our models.

On governance, watsonx.governance monitors the entire lifecycle of a model, from the first request through to production, and consolidates the facts on every model, whether built with IBM or third-party tools, into a single dashboard. It tracks which data is used and detects bias, which is exactly what is needed to comply with regulations such as the EU AI Act.

On control, the clients can run our stack on-premises or in their own environment rather than being forced into a public cloud. The models that we deploy are owned by the client. In regulated and sovereign contexts, that is decisive. Trustworthy AI isn’t a feature you bolt on at the end; it is the architecture.

TER: How is IBM helping organizations create a foundational structure that supports AI adoption? Can you share case studies of how IBM’s AI is transforming organizations and industries?

Vasques: One of the biggest challenges in AI adoption is data, not models. Our approach brings watsonx.ai, watsonx.data, and watsonx. governance into a single governed environment, organised around three pillars: data, AI, and governance. This enables an organisation to move from experimentation to production without friction, with the freedom to use Granite, open-source, or third-party models. And we do not use one model for everything. We build smaller, specialised models for specific domains such as HR, procurement, finance, or engineering. This method saves cost and energy. It also keeps the client’s data and intellectual property local. We build hybrid by design on Red Hat OpenShift, so AI runs where the data already lives.

A good example is L’Oréal. With around 4,000 researchers worldwide, they built a custom foundation model with us to replace non-renewable ingredients with more sustainable, bio-sourced ones, deployed in their own environment and fully aligned with their strategy. The lessons are consistent across projects: start by unifying and governing your data, choose models pragmatically rather than locking in, govern from day one, and measure ROI relentlessly.

TER: For organizations evaluating AI engineering platforms today, what criteria should they use beyond code generation capabilities, and what signals indicate a platform is ready for enterprise-scale AI adoption?

Vasques: Code generation is now table stakes. Specifically for development, we built IBM Bob, an agentic AI for large organisations tackling complex challenges: it does not just generate code, it works across the whole software lifecycle, taking into account architecture, data, security, and a company’s own standards.

But the real signals of enterprise readiness are elsewhere. First, governance and observability are built into the platform rather than bolted on later: can you trace, audit, and monitor every model and agent decision?

Second, model freedom: a serious platform routes work across Granite, open-source, and partner models without locking you to a single vendor. Third, integration with your real systems and data, because that is where the value lives.

Fourth, deployment portability and sovereignty: can it run on-premises, in your own cloud, or in a regulated EU region?

Fifth, agent lifecycle management as you scale from one agent to a fleet.

If a platform can govern many agents across a hybrid infrastructure with measurable ROI, it is enterprise-ready.

TER: The energy industry is undergoing rapid transformation in terms of decarbonization, grid modernization, and changing customer expectations. What is IBM doing in the energy sector?

Vasques: The energy system is being reshaped by three forces at once: decarbonization, grid modernization, and surging electricity demand, not least from AI itself.

Our role is to bring AI and automation to that transition without ever compromising operational reliability, on a hybrid and sovereign-capable architecture, which matters for critical national infrastructure.

We see the most value in three areas. First, intelligent asset management and predictive maintenance on critical infrastructure: energy companies can use IBM Maximo on their path to net zero or to manage critical assets.

Second, operational efficiency in renewables. We can optimise operations and data transparency, or use computer vision to improve wind-turbine blade manufacturing.

Third, sustainability and emissions intelligence, where companies can automate their ESG reporting.

Across all of these, the throughline is the same: ground AI in trusted operational data, forecast demand and renewable generation, optimise grid operations, and keep humans firmly in command of safety-critical decisions.

TER: Sustainability goals are now a boardroom priority. What are the competitive advantages of IBM Envizi in terms of emission management compared to other technologies in the market?

Vasques: Envizi’s edge is finance-grade auditability combined with breadth and automation. It automates the capture and consolidation of more than 500 ESG data types into a single, auditable system of record, and it calculates emissions across Scopes 1, 2, and 3 in an engine built on the GHG Protocol. Three things separate it from point tools: automated data capture from utility bills, interval meters, and renewable assets rather than spreadsheets; an Emissions API that embeds GHG calculations directly into a client’s existing workflows; and native integration with the rest of the IBM portfolio for asset and operations management.

TER: What changes do you foresee in the roles and responsibilities of AI working alongside the human workforce over the next few years, and what is your outlook for AI in 5-10 years?

Vasques: Over the next few years, the pattern is augmentation, not replacement. Agents will absorb routine, high-volume work, while people move toward judgment, creativity, relationships, and oversight. New roles will grow around governing AI: supervising agents, auditing their decisions, and managing risk.

Governance is not optional here, especially in Europe. With the EU AI Act, we need to know who is using which model, for what, and be able to prove it. That is what keeps us in control as adoption scales.

Within five to ten years, I expect the agentic enterprise to become the norm. At Think 2026, we introduced capabilities such as IBM Sovereign Core for operational independence and the next generation of watsonx Orchestrate as an agentic control plane, precisely because organisations need to run AI at scale across many agents, with governance and sovereignty built in. The throughline is always the same: powerful AI, deployed responsibly, with humans firmly in command.