After three decades in enterprise software, I've seen enough hype cycles to distinguish when a technology is genuinely disruptive and when it's merely a good cover story. AI agents fall in the first category — and most business leaders haven't yet calibrated the magnitude of the change.
The most common conceptual mistake
Almost every conversation with CEOs and boards about AI agents starts with the same confusion: treating them as a more sophisticated version of chatbots. They are not. A chatbot answers questions within a script. An agent, by contrast, is an autonomous system that pursues a goal, makes sequential decisions, accesses tools, executes actions on real systems and learns from the result.
The difference isn't capability — it's architecture. And that changes everything: how the business case is designed, what risks need mitigation, how ROI is measured, and above all, what kind of governance it requires.
Three high-value applications in banking and insurance
In the next twelve to twenty-four months we'll see production deployments —not pilots— in at least three fronts:
Underwriting and risk assessment in insurance. Agents capable of processing heterogeneous documentation, querying internal and external databases, executing regulatory rules and issuing underwriting recommendations in minutes. The human analyst becomes an exception reviewer, not a file processor.
Collections and portfolio restructuring. Agents that negotiate payment plans on digital channels, evaluate payment capacity in real time, escalate only complex cases to humans, and learn from successful negotiations. Collections moves from a contact-center process to one governed by intelligent systems.
Onboarding and assisted KYC. Agents that orchestrate identity verification, documentary validation, list screening, PEP evaluation and alert issuance — maintaining complete traceability for the regulator. It's the next frontier of compliance.
What the board should ask before approving investment
The most expensive mistake I see in boardrooms isn't investing in AI — it's approving AI investment without asking the right questions. These are the five a well-prepared board should be able to answer before signing:
1. Where exactly is the ROI? Not "operational efficiency" as a general concept — the concrete, measurable, attributable metric. If the answer is vague, the project will die in pilot.
2. How do we govern autonomous decisions? An agent can execute a thousand decisions per hour. What thresholds require human approval? How do we audit what was decided and why? Who is responsible when it's wrong?
3. What regulatory risks are we assuming? In Mexico, the CNBV and CONDUSEF are building criteria. In the EU, the AI Act is already in force. An agent deployed without a governance framework can become a regulatory liability greater than any operational savings it generates.
4. Are we building or buying? Building foundational technology is increasingly less relevant — but competitive differentiation lies in how it's composed, integrated and trained with proprietary data. Middle-layer strategy (orchestration, data, observability) is the most important decision.
5. Do we have the talent to sustain this? An agent deployed without a team capable of operating, monitoring and improving it degrades quickly. The most underestimated investment is in the operating model.
The mental shift coming
For 30 years we've thought of analytics as something that produces reports for humans to make decisions on. Agents invert that logic: the system decides and acts, and the human supervises, audits and corrects. It's a deep operational paradigm shift — deeper than cloud, deeper than mobile. Organizations that understand it early and govern it well will have a competitive advantage that's hard to reverse.
The right question for 2026 is no longer "should we use AI?" — that discussion is closed. The right question is: "what business decisions are we willing to delegate to a system, and how do we govern them?"
If your board is still discussing the first, you're late. If they're seriously discussing the second, you're in the right place.