Most conversations about agentic AI focus on how autonomous these systems can become. Financial services leaders are asking a different question: Where should autonomy stop?
For banks and insurers, the challenge is not simply deploying AI agents. It is determining how those systems, governance requirements and human expertise can work together in decisions that carry financial consequences.
During a recent discussion among leaders from JPMorgan Chase, Allianz, AWS and SAS, one theme emerged repeatedly: the future of agentic AI is not about removing humans from decisions. It is about redefining their role.
From human in the loop to human in the lead
For years, discussions around AI governance have centered on keeping a human "in the loop." But some organizations are beginning to view that relationship differently.
As Andrea Pohlman of Allianz put it:
"They've stopped using human in the loop. It's human in the lead."
The distinction may seem subtle, but it reflects a meaningful shift in how financial institutions are approaching AI adoption.
Rather than asking where AI can replace people, organizations are evaluating where AI can support them. Agentic AI can gather information, summarize findings, identify patterns and recommend next actions, but accountability for important outcomes remains with humans.
That approach reflects the realities of banking and insurance, where decisions often carry financial, regulatory and reputational consequences. While AI can help accelerate processes, institutions still need people to evaluate risk, exercise judgment and maintain trust with customers.
Where agentic AI is creating value today
The strongest opportunities for agentic AI are not fully autonomous systems. Instead, many organizations are finding value in operational processes that are repetitive, time-consuming and highly manual.
At JPMorgan Chase, Adolfo Lopez described many current agentic AI applications as "assistant or delegative," helping employees enhance processes and support decision-making rather than acting independently.
As a result, many organizations view KYC and customer onboarding as practical areas for agentic AI support. AI agents can help collect information, identify missing documentation and automate portions of the workflow, allowing employees to spend less time on administrative work and more time serving customers.
As AWS's Sri Raghavan noted, "a lot of manual tasks are done as a part of the KYC process."
Insurance organizations are seeing similar opportunities in claims processing. Agentic AI can help gather information, review documentation and route cases more efficiently, allowing claims professionals to focus on customer interactions and more complex decisions. Allianz has already seen early success using AI in high-volume claims scenarios where speed and efficiency are critical. In these cases, AI can accelerate the process, but humans remain responsible for the ultimate decision.
Across these use cases, the goal is not to remove human involvement. It is to reduce administrative burden so employees can focus on higher-value work.
Why human oversight still matters
If these use cases are proving valuable, why aren't organizations moving more quickly toward autonomous decision-making?
The challenge becomes clearer in high-stakes decisions. Consider a hypothetical underwriting workflow. An AI agent could collect applicant information, cross-reference supporting documents and generate a recommended risk assessment. But because lending decisions carry significant financial and regulatory implications, organizations still need clear oversight, explainability and accountability for the final outcome.
According to the panelists, the biggest obstacle may not be technology.
As Lopez explained:
"The technology is there. But it's the regulations that constrain us."
Financial institutions operate in some of the most heavily regulated environments in the world. Activities such as lending, underwriting, claims processing and risk management are subject to strict requirements for fairness, transparency and accountability. An incorrect decision can create legal, financial and reputational consequences.
Those realities are shaping how organizations deploy agentic AI. While many institutions are comfortable using AI to gather information, analyze data and support decisions, they remain cautious about allowing AI systems to independently make consequential choices.
Pohlman pointed to insurance as an example. While AI can synthesize information from numerous sources, she emphasized the importance of maintaining human responsibility for risk-related decisions, particularly in complex cases.
In other words, the challenge is not whether AI can contribute to decisions. It is deciding which decisions still require human judgment and accountability.
A new operating model for agentic AI
The institutions that succeed with agentic AI may not be the ones that pursue the highest level of autonomy. They may be the ones that most effectively combine intelligent automation with human expertise, governance and accountability.
The question is not whether humans stay involved. It is how their role evolves as confidence in AI systems grows.
As organizations continue to explore agentic AI, the most effective operating model may be the one Pohlman described: not human in the loop, but human in the lead.
In financial services, the question is not how much autonomy AI can achieve. It’s where autonomy creates value and where human judgment must remain.
