
AI agents, governance, adoption, data, and value measurement shaped the conversation at FEBRABAN TECH 2026, one of the region’s leading events for financial innovation.
From August 24 to 26, FEBRABAN TECH 2026 brought together leaders from banking, fintech, and the technology industry at Distrito Anhembi to discuss how to move Artificial Intelligence from experimentation into real-world processes.
Organized by the Brazilian Federation of Banks, the event is one of the leading forums for tracking the evolution of technology applied to financial services. This year’s central theme, “Smart agents, human leadership”, focused on a new stage: systems capable not only of assisting, but also of executing tasks, connecting with data, and participating in business processes.
For Latin America, these conversations are especially relevant in a context shaped by the growth of digital banking, fintech, new payment methods, and Open Finance. The discussion now includes a new challenge: how to integrate AI into this evolution and bring it into production within financial organizations.
From assistants to AI agents
One of the topics discussed at FEBRABAN TECH 2026 was the evolution from assistants to AI agents.
While an assistant typically responds to a request, analyzes information, or generates content, an agent can execute actions within a process and coordinate with other systems.
This opens the possibility of working with networks of specialized agents, each with specific functions and connected to data, applications, and business rules.
The discussion was not limited to the technical capabilities of these agents. It also raised questions about governance, integration, security, and human oversight.
GenAI in production: governing from the center while enabling innovation
Daniel de Amaral · Managing Director · Accenture
Led one of the panels focused on generative AI in production, introducing the different topics of discussion.
One of the ideas that ran throughout the panel was the need to centralize governance and decentralize innovation.
In regulated organizations, strategy requires clear guidelines and centralized direction. At the same time, individual teams need room to experiment and identify concrete applications of AI within their own processes.
The panel also highlighted the importance of working with partners that can support the transformation process and help sustain implementation in organizations where change often requires coordination across multiple areas.
Another shared point was that the main obstacle is not always technological. Organizational culture and change management emerged as key factors in moving from pilots to real adoption.
Risk, cultural change, and customer experience
Cassio Pereira · Partner and Director of Data & AI for Business · Itaú Unibanco
Addressed AI adoption from the perspectives of risk, organizational transformation, and customer relationships.
He noted that moving from a traditional model of change to a more accelerated one requires organizations to rethink how risk is managed.
He also pointed out that AI can generate incremental value across many processes, but capturing that value depends not only on the technology itself, but also on effectively managing cultural change.
When discussing differentiation, he focused on customer understanding. In his view, the advantage will belong to those who can understand customers better and use hyperconnected platforms to deliver more relevant products and experiences.
Adoption, experimentation, and agentic infrastructure
Thiago Suzano · Director of Technology – Engineering, Data & AI · B3
Shared a concrete experience involving internal AI adoption.
According to Suzano, a 32-hour large-scale training initiative did not initially produce the expected results. Real learning began when people started experimenting with the technology and applying it to concrete situations.
Based on that experience, the team moved toward a combination of agentic infrastructure, data culture, and smaller models designed for specific organizational problems, with the goal of generating measurable impact on results.
Suzano also highlighted how the boundaries between industries are changing: a financial institution can develop technology with enough value that it ultimately becomes something it can commercialize.
Global scale and local adaptation
Julio Blanco Perez · CTO · Santander Brasil
Shared the perspective of a global group with more than 140 million customers and operations in countries at different stages of transformation.
He emphasized the need to combine global capabilities with the realities of each market.
Within this approach, he mentioned the use of platforms as a service to accelerate adoption, supported by local teams capable of adapting those platforms to the specific needs of each country and sustaining growth.
“Apply, solve, and go to market.”
He also discussed the group’s commitment to developing its own agents and mentioned the use of Devin as one of the tools they are currently working with.
When discussing the future of banking, he summarized his vision in one word: simplification.
Specialized models for different needs
Leandro Marçal Araújo · Director of Technology · Banco Bradesco
Addressed AI model architecture and the value of avoiding a single solution for every use case.
His approach was to use multiple specialized LLMs depending on the product or agent, rather than forcing every process through a single model.
This approach allows organizations to assign different models according to the function each solution needs to perform, while maintaining a common governance framework.
From cost per token to cost per task
Marcelo Alvarenga · GTM, LATAM · Anthropic
Proposed one of the panel’s most concrete approaches: moving beyond measuring only cost per token and beginning to measure cost per task.
The idea is to compare how much it costs to complete a process before and after introducing AI.
If a task that previously carried a high cost can be completed for a fraction of that amount through an agent, value begins to be measured by the outcome of the task rather than solely by the model’s technical consumption.
Alvarenga also used the framework of People, Processes, and Product to discuss adoption.
Within People, he distinguished between two groups: those who use the tools within the organization and the customers who receive the value generated by those tools.
As a possible adoption path, he suggested starting with people by creating teams of champions, then reviewing processes in other areas and asking how they could be addressed with AI, and ultimately advancing until those improvements are reflected in the product or service delivered to the customer.
He also highlighted two characteristics of AI-native players: speed of execution and the ability to think AI-first even when their data is not yet perfectly organized.
He also pointed to the growing potential of tools capable of autonomously executing a significant portion of the work.
Security, identity, and governance
In addition to panels focused on adoption and production, FEBRABAN TECH 2026 included conversations around security, digital identity, and AI governance.
The growth of agents and autonomous systems increases the need to establish clear rules around data access, traceability, consent, fraud, and human oversight.
As AI moves from generating information to executing actions, these factors become increasingly relevant within the architecture of each solution.
This year’s theme, “Smart agents, human leadership”, captured this balance between greater technological autonomy and human responsibility.
Applications in sales processes
Although many of the conversations at FEBRABAN TECH originate in the financial sector, the same principles can be applied to other areas of business.
In sales processes, for example, AI agents can participate in tasks such as data enrichment, lead outreach, qualification, interaction logging, and follow-up.
At InterConnecta, for example, this evolution is being applied through DialRun and AI Sales Ops agents, connecting different stages of the sales process.
An agent can enrich the available information about a lead, start a conversation, identify interest, log the outcome, and trigger a follow-up by email or WhatsApp. When an opportunity is identified, the process can continue with a member of the sales team.
Financial AI in Latin America enters a new phase
The conversations at FEBRABAN TECH 2026 presented different perspectives on the same challenge: how to move AI from pilots into operations.
The topics discussed included governance, cultural change, value measurement, model specialization, local adaptation, agentic infrastructure, customer experience, and internal adoption.
For Latin America, these discussions build on a transformation already underway through digital banking, fintech, new payment methods, and Open Finance.
The next phase will be shaped by how institutions integrate AI into real-world processes while balancing innovation, security, governance, and business outcomes.


