Last week I had the privilege of speaking at the Cari AI Summit 2025, where I shared one of the topics I’m most passionate about: how AI truly moves from design to operation in the enterprise.

As someone who has spent nearly three decades building and leading technology solutions across industries, from telco and banking to healthcare and digital transformation, I’ve seen firsthand that AI is not just a model you deploy. It’s a living system. And like any living system, it requires care, governance, architecture, and above all, responsibility.

Too often, the conversation around AI is shiny and front-facing: a new chatbot here, a new model announcement there. But behind the curtain, enterprises struggle with questions that don’t always make it into the headlines:

• How do we make sure our data foundation is trustworthy?
• Who is accountable when an AI system behaves unfairly?
• How do we embed AI into existing business processes rather than leave it as a side experiment?
• How do we balance innovation with compliance, especially in regulated industries?

These are the backstage realities of AI. And they’re the ones that ultimately define whether a project succeeds or fails.

Designing AI for the Enterprise

The most common trap I see companies fall into is starting with the “what” before the “why.” They rush to adopt the latest tool or API without aligning it to a business problem or cultural need.

AI initiatives must start with problem framing, understanding what matters to the business and why AI is the right tool. From there, data becomes the true foundation. High-quality, governed data pipelines aren’t glamorous, but they make or break enterprise AI projects.

And then there’s Responsible AI. For me, explainability, fairness, and compliance aren’t add-ons, they’re design principles. If your AI cannot be explained, trusted, or regulated, it doesn’t belong in production.

Finally, architecture matters. Hybrid cloud, multi-platform designs, and scalable patterns separate proof-of-concepts from production systems. The best enterprises think about architecture from day one.

Building the AI Factory

Think of this stage as building the factory that produces AI at scale.

We’re no longer just training a single model; we’re managing pipelines. MLOps is the DevOps of AI , handling model lifecycle, monitoring, retraining, and deployment.

Security and compliance run deep here: securing models, protecting sensitive data, and aligning with sector-specific regulations (in my world, often banking, healthcare, and telco).

And let’s not forget the people side. AI changes the roles of teams: data scientists, engineers, architects, and critically business translators who ensure AI stays relevant to business needs.

Operating AI at Scale

AI isn’t something you “finish.” It’s a continuous loop. Models drift. Bias emerges. Regulations change. Business priorities shift. That’s why monitoring, feedback, and retraining are so critical.

Integration is another big hurdle. AI is valuable only when it’s embedded in real workflows, not when it sits in a lab or innovation hub.

At the same time, cost is always on the table. I push leaders to shift the narrative: it’s not about “AI is expensive.” It’s about “AI creates measurable business impact.” When you can demonstrate that link, the conversation changes.

The Invisible Challenges

What makes enterprise AI truly complex isn’t always the tech. It’s the human and cultural side:

• Building trust among executives and employees.
• Navigating the scarcity of talent, while upskilling teams already in place.
• Operating under regulatory fog, especially with frameworks like GDPR and the upcoming AI Act.
• Dispelling the myth that AI is “plug and play.”

These invisible challenges are what separate enterprises that thrive with AI from those that stall out.

Looking Ahead

The next decade won’t be defined by AI hype, it will be defined by responsible automation.

We’ll see more autonomous systems balanced with human-in-the-loop governance. AI will move beyond isolated clouds into multi-cloud and distributed edge ecosystems, even in the most regulated industries.

And if there’s one lesson I carry from my 28+ years in technology, it’s this: AI isn’t magic. It’s the result of disciplined design, relentless operations, and responsible leadership.

Enterprises that master both the vision and the backstage execution will own the future.

Behind the scenes, AI isn’t just a technology, it’s a leadership challenge. And those who get it right will shape the next decade of enterprise transformation.

— Ricardo