While humans make the final decisions, Agentic AI handles the validation and analysis.
Insurers are beginning to use agent-based artificial intelligence (AI) to automate underwriting, document verification, and customer service, but decisions that involve financial and regulatory risk still require human oversight.
speak at Asian banking and finance and insurance asia Summit in Singapore, APAC Director of German IT consulting and software solutions company valantic Gonzalo Carvalho Insurers say they are moving beyond AI assistants to AI agents that can independently run some of their business processes, especially in underwriting workflows.
Carvalho said that as AI adoption accelerates, insurers also need to build governance into these systems to reduce illusions and maintain auditability.
“AI, you know, you have to use the right context and guardrails, which is very important, but also difficult to structure properly,” he said.
He noted that while many technology leaders remain cautious, companies are expected to significantly increase AI projects in the coming years.
“By 2028, we expect to see a 2,500% increase in projects using AI,” Carvalho said. At the same time, “67% of IT leaders are currently dissatisfied with AI.”
He added that AI-assisted software development is also becoming the norm, with “90% of engineers” expected to use AI-assisted development tools and “55% of developers” running multiple coding tools simultaneously.
Carvalho said increased productivity also creates new challenges for governance.
“AI has made development much easier, but the key question is: Are we building solutions that are safe, secure, scalable, and well-managed?” he said.
To illustrate how agent AI is being applied to insurance, Carvalho demonstrated a life insurance workflow where an AI agent verifies customer ID, proof of address, and medical information before routing an application for underwriting.
AI also generated risk ratings, allowing insurers to query the reasoning behind recommendations through a chatbot interface.
During the demo, the AI rated the applicant’s risk at “68%.” Acting as a human underwriter, Mr. Carvalho ignored the recommendations and approved the application.
“I choose to approve it because I am a human being and I have the ability and authority to do it,” he said. “That depends on my ability to approve, not yours.”
This workflow also generated offer letters, customer emails, and audit logs to record whether an action was performed by an AI agent or a human user.
Carvalho said that while agent AI can automate repetitive insurance tasks, organizations still need clear governance of how AI systems validate information, generate recommendations and record decisions.
“Agents already perform proof of address, ID, and other verifications that may be required,” he said, but added that human reviewers are ultimately responsible for the underwriting decision.
