While the public imagination has been captivated by the conversational abilities of chatbots, a new report from McKinsey & Company suggests the global banking industry is quietly approaching a far more profound transformation: the era of agentic AI.
According to the report, the banking sector is shifting from a period of broad experimentation to a paradigm shift defined by autonomous agents, systems capable of planning, executing multi-step workflows, and using tools with minimal human intervention. This transition represents the move from “hype” to “precision.” It is no longer about the novelty of a machine that can write a poem; it is about the utility of a system that can autonomously reconcile a ledger or migrate a mortgage.
However, this technological leap comes with a stark warning. McKinsey predicts that while agentic AI could reduce bank unit costs by 15 to 20%, it also threatens to erode up to $170 billion in global profit pools by 2030 if banks fail to adapt their business models.
For more like this on Forbes, What Is Agentic AI And What Will It Mean For Financial Services?
To understand how financial institutions are bridging the gap between theoretical potential and production-grade deployment, I spoke with Jonathan Pelosi, Head of Financial Services at Anthropic, Scott Mullins, Managing Director of Financial Services at Amazon Web Services and Steve Suarez, CEO of HorizonX, Senior Advisor to McKinsey and former Global Head of Innovation, GF at HSBC.
Video: Jonathan Pelosi, Head of Financial Services at Anthropic
The 2026 Trust Horizon
For years, the adoption of AI in banking has been throttled by a trust gap. In a regulated industry, a model that hallucinates facts is a serious liability. Pelosi argues that this gap is closing rapidly due to the evolution of evaluation frameworks.
“A year ago, when [researchers] checked, there might be 8 out of 10 facts that were correct,” Pelosi said. “Now you’re getting like 99 out of a 100.”
Pelosi identifies 2026 as the year the industry reaches a psychological and statistical tipping point. He draws a parallel to the adoption of autonomous vehicles. Just as passengers need data to trust a driverless car, bankers need data to trust an agent.
“Going from 80% to 99% accuracy is impressive, unless you are a bank,” said Suarez and added, “At 1% error a system still misreports 100 balances out of 10,000. AI in finance must aim for near-zero mistakes.”
Video: Scott Mullins, Managing Director of Financial Services at Amazon Web Services
Beyond AI Tourism
As the technology matures, the industry’s approach to implementation is maturing with it. Mullins observes that banks are moving away from AI tourism, running pilots simply to claim they are innovating.
“If what you’re trying to accomplish is simply, ‘I want to do an artificial intelligence experiment,’ that’s not really a true business outcome,” Mullins said. “Where people are seeing the most value is in having a very specific business outcome in mind.”
This shift from “wow” to “how” is driving banks toward what Pelosi calls the “unsexy stuff.” The most impactful immediate use cases are not flashy chatbots, but deep operational improvements in the mid and back office.
For more like this on Forbes, The Legacy Banks Quietly Building The Future Of Finance.
The ‘Unsexy’ Revolution
One of the most critical applications for agentic AI is modernizing the industry’s aging infrastructure. Many financial institutions still rely on COBOL-based systems written decades ago.
“It turns out these institutions are built on 40, 30 year old legacy code that honestly people no longer know how to even code,” Pelosi said.
He noted that Anthropic’s models are now successfully modernizing this legacy code, effectively reading millions of lines of archaic programming and refactoring it into modern languages.
Similarly, compliance workflows like KYC are moving from human-heavy processes to agent-led automation. Mullins points to compliance reporting and risk management as areas where agents can significantly reduce manual intervention while improving accuracy.
However, integrating these agents requires navigating the reality of “repairing a flight whilst flying it,” as Mullins described it. Banks cannot shut down core systems to upgrade them; they must integrate AI agents into live, mission-critical environments.
The Disruption Threat: The Shopping Agent
While banks focus on internal efficiency, the McKinsey report highlights a significant external threat: the rise of the shopping agent.
Historically, banks have profited from customer inertia. It was simply too difficult for consumers to constantly switch accounts to find the best yield. Agentic AI is poised to remove that friction. McKinsey predicts that consumer-facing AI agents will soon be able to autonomously monitor interest rates and move deposits to top-of-market accounts.
If just 5 to 10% of checking balances migrated to higher-yielding accounts driven by these agents, the industry’s deposit profits could decline by 20 percent or more. This trend forces banks to compete not just with other banks, but with the algorithms managing their customers’ financial lives.
Governance: The Human-in-the-Loop
To navigate some of the risks, both Pelosi and Mullins emphasize the necessity of human-in-the-loop governance. The goal is not to replace the banker but to sandwich the AI agent between layers of human oversight.
“You still have the added benefit of while it can do 80-90 percent of the heavy lift, the humans are still very much in the loop to make sure that the checks and balances are in place,” Pelosi said.
Mullins advises CIOs to adopt a “golf bag” approach to this technology, utilizing different models for different tasks rather than relying on a single vendor. This allows banks to select the most secure and precise tools for specific workflows, ensuring that governance evolves alongside the technology.
Key Takeaways for Bank Executives
1. Target the ‘Unsexy’ for High Impact
Stop chasing novelty. Direct investment toward “unsexy” mid-office bottlenecks such as legacy code modernization and automated compliance reporting. These areas offer the clearest path to the 15 to 20% cost reductions predicted by McKinsey.
2. Prepare for Algorithmic Competition
Recognize that customer inertia is ending. As shopping agents begin to automate switching, banks must move from broad segmentation to a “segment of one.” Use internal AI to proactively offer hyper-personalized value to customers before an external agent moves their money elsewhere.
3. Operationalize ‘Sandwich’ Governance
Do not deploy agents without upgrading supervisory procedures. Implement a workflow where humans define the goals and validate the outputs, while agents handle the execution. As Mullins warned, simply putting an agent onto a workflow without adjusting human supervision is a recipe for failure.