According to recent theCUBE Research analysis, 93% of enterprises are developing or piloting custom AI agents, yet only 21% have established mature governance, data quality, and integration frameworks to support them at scale. As agentic AI adoption accelerates, enterprises are discovering that autonomy without structure quickly becomes risk, not advantage.
In this episode of AppDevANGLE, I spoke with Tiago Azevedo, Chief Information Officer at OutSystems, about how CIOs should think about agentic AI beyond experimentation and what it takes to operationalize agents in a way that delivers measurable business outcomes without sacrificing trust, data quality, or architectural integrity.
Our conversation explored why governance must evolve alongside autonomy, how poor data foundations are being exposed by AI agents, and why API-first, composable architectures are becoming a prerequisite (not an optimization) for agentic AI success.
Strategic Guardrails Matter More Than Speed
One of the most consistent mistakes enterprises make with agentic AI is chasing speed without strategic alignment. A proliferation of agents does not automatically translate into business value.
As Tiago put it: “The whole topic of IT strategy in the past was supposed to be linked to the company’s strategy. The same really applies now for agentic AI—not just going after the buzz of a gold rush, but actually picking the right use cases that bring impact.”
Rather than deploying agents everywhere, OutSystems focuses on identifying domains where reasoning, context awareness, and autonomy can create meaningful breakthroughs. This framing shifts agentic AI from an efficiency tool into a strategic capability that is explicitly tied to growth, margin, and operational performance.
AI Is Exposing Data Debt, Not Creating It
While governance is often discussed as a policy problem, Tiago was clear that data quality is the true constraint on agentic AI effectiveness.
“AI has been able to surface the importance of data. If your data wasn’t ready before and you’re only realizing that now, then companies will have to go back and clean it. AI is surfacing gaps that already existed.”
Agentic systems rely on continuous access to operational, customer, and transactional data. When those foundations are fragmented, inconsistent, or poorly governed, agents amplify the problem rather than solve it.
This reality explains why only a small percentage of enterprises trust agents to make or recommend critical decisions today, and why governance cannot be deferred until “later.”
Governance Must Travel With the Agent
Unlike traditional automation, agentic AI systems can traverse multiple systems, data domains, and workflows autonomously. That makes static, centralized governance models insufficient.
Tiago emphasized the need for embedded guardrails: “Agents through tools are able to access multiple sets of data in the company. You have to put guardrails behind it, otherwise anyone can build an agent that accesses confidential information and uses it improperly.”
The implication is clear: governance must be enforced at the architectural and platform level, not left to individual teams or post-deployment audits. Human-in-the-loop controls remain critical, but they must be designed as part of the system instead of being layered on after the fact.
API-First Architecture Is the Enabler, Not the Bottleneck
Integration complexity remains one of the top blockers to agentic AI adoption, cited by 64% of enterprises in recent research. Tiago sees this less as an AI problem and more as an architectural reckoning.
“If you have a composable, modular, API-first architecture, when you add agents on top everything works smoothly. If you don’t, you’re just adding agentic AI on top of a messy foundation.”
Agentic AI does not replace the need for clean enterprise architecture; it intensifies it. Organizations that invested early in service-oriented and composable design are finding it far easier to layer agentic intelligence across their environments without disrupting business continuity.
From Automation to Autonomy
Perhaps the most important distinction Tiago drew was between automation and agentic AI.
“Traditional automation was about efficiency—making manual tasks faster. Agentic AI is about autonomy. These systems reason about goals, context, and intent. That’s what unlocks entirely new levels of impact.”
At OutSystems, this shift is already visible in production use cases across sales, customer support, and finance, where agents reduce decision latency by up to 40% by reasoning across telemetry, CRM data, knowledge bases, and financial systems.
This is not incremental improvement. It is a structural change in how decisions are made and work is executed.
Analyst Take
Agentic AI is forcing enterprises to confront long-standing weaknesses in data quality, integration, and governance. The technology itself is not the limiting factor—organizational readiness is.
What stood out in this conversation is the clarity of sequencing: strategy first, data second, architecture third, autonomy last. Enterprises that reverse that order may achieve short-term gains, but they will struggle to scale trust and impact.
Agentic AI is not simply the next phase of automation. It is an operating model shift that rewards disciplined architecture, measurable outcomes, and governance that travels with intelligence. Organizations that internalize this early will define the next generation of enterprise software delivery.
