Agentic AI
What Enterprises Really Asked About AI Agents at EY Lisbon
Enterprises want AI, but not at the cost of control. Discover what they really asked and why architecture matters more than ever.
A couple of weeks ago, Hypatos joined EY in Lisbon for a closed-door roundtable on AI in enterprise finance and operations. We were one of only two vendors invited to present - a chance not just to demo, but to listen.
The room was filled with transformation, finance, and IT leaders from some of Portugal’s largest enterprises. Their use cases varied, but one message cut across roles and industries: AI is overdue.
And yet, no one felt confident about where to start or how to start without risking regret.
Excitement and caution tend to travel together at these events. And nowhere was that clearer than during our demo sessions on tax compliance.
Every enterprise needs to modernize how tax logic is handled. But few are comfortable being first to the party. Why? Because tax isn’t just data. It’s liability. Once AI enters that equation, tracking back the way it makes decisions has to be guaranteed or it becomes the risk.
The question in the room wasn’t “Should we use AI?”
It was: “Can we use AI --and-- not lose control?”
We showcased a concrete, high-impact, high-complexity use case: automated tax validation. Delivered through our Invoice Processing AI Agent, it tackled a task every finance team knows - applying local VAT rules, flagging edge cases, and handling escalation with minimal effort.
What resonated most in the room was how the AI Agent felt like an automation solution that wouldn’t become a maintenance burden or a regulatory risk. And it wasn’t just the outcome that counted.
The Q&A put something else into focus. When teams ask how easy it would be to fix, change, or audit later, what they were actually asking is if control and visibility would be traded away for AI-powered speed.
And that was our a-ha moment - the conversation had moved from results to what's under the hood. Teams weren’t worried about features. They were asking us, in business language, whether the architecture was a black box.
It’s a fair concern. When AI solutions are built on traditional SaaS stacks, components are tightly coupled, logic is buried inside models, and even minor changes require retraining. Locating the root of an issue or updating logic without disrupting the system can be nearly impossible.
Agentic architecture, the one Hypatos offers, changes the game. It is built for transparency and explainability where logic is visible in prompts and the agent explains every decision it makes. That’s what makes it a safer starting point for enterprise AI, especially in domains where mistakes are costly and control is non-negotiable.
Teams enter demo conversations with us asking for help with invoice coding, tax validation, or reconciliation. And that’s exactly what our agents are built for.
But the moment the conversation turns to auditability, ERP fit, or evolving logic, they’ve moved into architecture territory, whether they know it or not. This happens 10 out of 10 times, and for good reason. Architecture is where long-term control is either won or lost.
We believe the topic deserves a business-first exploration, so, we are launching a pre-summer miniseries of three articles where we’ll unpack the architecture questions every GBS buyer should be aware of.
Don't miss them - subscribe to our LinkedIn newsletter!
What we took away from the roundtable was that enterprise buyers, the ones in Portugal at least, are not lost or overwhelmed. It’s that they’re deliberate. Behind every “we’re not sure how to start” is a real requirement: keep us fast, but keep us in control. They’re not pushing back against AI - they’re holding out for the kind that doesn’t compromise traceability, flexibility, or independence. And once the architecture supports that, the rest of the decision gets easy.
Ready for agentic process automation? Check the free tools we've created to help you make the right decision and innovate with confidence:
Further stories from our blog
.jpg)
