The AI Agent Gap

The global AI banking market hit $20.6 billion in 2026. That number gets thrown around a lot, but what matters more is where that money is going. And where it is not going yet. The gap between “we use AI” and “we have autonomous agents in production” is where the real opportunity lives right now.

The Numbers

Here is the adoption spectrum as it stands in early 2026. McKinsey’s latest survey puts overall AI adoption at 88% of organizations. That is the highest it has ever been, up from 72% in 2024. NVIDIA’s State of AI in Financial Services report shows 42% of firms are actively exploring agentic AI. Deloitte’s banking survey finds only 37% are training staff properly on AI tools. And the number that matters most: just 21% have agentic AI running in production (NVIDIA 2026).

That last number is the one worth sitting with. Almost everyone is using AI. Less than a quarter have agents that can act autonomously within defined guardrails.

The spread tells a story. Adoption is near-universal. Experimentation is growing fast. But production deployment of autonomous agents is still early. The firms that close this gap first will define the next generation of financial services.

That bottom bar is the quiet one. According to Accenture’s 2026 Banking Technology Vision, only 12% of financial institutions have a well-defined, enterprise-wide AI agent strategy. Plenty of banks have AI strategies. Far fewer have agent strategies. The distinction matters because agents require a fundamentally different approach to governance, risk management, and human oversight.

Who Is Building

The institutions making real moves on agentic AI share a common trait: they committed early and are iterating fast. This is not about which bank is “winning.” Every institution below is solving a different version of the same problem, and the variety of approaches is what makes this moment interesting.

Goldman Sachs partnered with Anthropic in late 2025 to build internal AI agents for due diligence, transaction review, and wealth management workflows. Their approach focuses on high-value, high-complexity tasks where an agent can reduce review time from days to hours while keeping a human in the loop for final decisions.

JPMorgan has been building their own large language models internally and deploying AI across trading, research, and operations. Their LLM Suite reportedly processes over a million queries daily from employees. They are investing in agents for trade execution optimization and regulatory compliance monitoring.

In Canada, the picture is especially compelling. CIBC’s CAI platform, which I covered in detail in my post on Canadian banks and AI, went from pilot to 20,000+ employees in eight months. That speed of deployment is rare in banking. TD Bank has been vocal about their push into agentic AI, with leadership specifically calling out autonomous systems as the next phase of their AI strategy. RBC’s Borealis AI lab continues to be one of the most productive financial AI research groups in the world.

The pattern across these institutions is consistent. Start with copilot-style tools that assist employees. Measure the impact. Then gradually expand toward agents that can execute multi-step workflows with appropriate guardrails and human checkpoints.

The Canadian Advantage

Canada ranks first globally for AI maturity in financial services according to Deloitte’s 2025 Global AI in Banking report. That ranking reflects something real. The Big Five banks collectively employ thousands of AI specialists, operate dedicated research labs, and are deploying at a pace that matches or exceeds their American counterparts relative to size.

What makes Canada’s position unique is the combination of strong regulatory frameworks, concentrated banking infrastructure (five banks serve the majority of the population), and a deep talent pipeline from universities like Toronto, Waterloo, Montreal, and Alberta. Those four cities alone have produced foundational AI research that the entire industry builds on.

I wrote about each of the Big Five and beyond in my breakdown of Canada’s top banks and AI. Here is where they stand on AI strategy:

Each bank has optimized for different strengths. RBC went deep on research. CIBC moved fast on deployment. TD is betting on agents. BMO is augmenting human expertise. Scotiabank is solving multi-market complexity. All of them are investing heavily, and all of them are producing measurable results.

The concentrated nature of Canadian banking means these five institutions can move an entire national financial system forward faster than the more fragmented American or European markets. When one bank ships something that works, the others notice and respond within quarters, not years.

The Training Factor

The institutions that invest in training their people on AI tools see significantly higher returns. That is the clearest signal in the data. Deloitte’s 2025 report found that firms with structured AI training programs reported 35% average ROI on their AI investments, compared to 12% for firms that deployed tools without dedicated training.

Only 37% of financial institutions are training staff properly on AI tools (Deloitte 2025). That number is growing, up from 28% in 2024. But it remains the single biggest predictor of whether an AI deployment succeeds or stalls.

The gap between those two numbers represents the largest opportunity in banking AI right now. Tools without training create shelfware. Training without the right tools creates frustration. The institutions that pair both together are the ones posting the numbers worth paying attention to.

CIBC’s approach is instructive. When they rolled out CAI, they did not just give people access and hope for the best. They built structured onboarding programs, created internal champions in each business unit, and measured adoption at a granular level. The result was 600,000+ hours of productivity gained within the first year. That is what happens when deployment and training move together.

The same principle applies to agentic AI. Agents are more capable than copilot tools, but they also require more sophisticated human oversight. The people supervising agents need to understand what the agent can do, where its boundaries are, and when to intervene. That requires training that goes beyond “here is how to use the tool” and into “here is how to work alongside an autonomous system.”

Banks that build this capability now will have a structural advantage as agents become more prevalent. The organizations that know how to deploy, govern, and collaborate with agents will move faster than those still figuring out the basics.

What Comes Next

The market is projected to reach $67.74 billion by 2030 according to Grand View Research. That is a compound annual growth rate of roughly 34%. But the number itself matters less than what it represents: every major financial institution on the planet is increasing AI investment simultaneously.

Here is what the next four years look like based on current trajectories.

Agentic AI moves from pilot to production across the industry. The 21% figure for agents in production will likely double by 2028 as the early movers demonstrate clear ROI and the platforms mature. The firms exploring agents today (that 42%) will transition to deployment as tooling, governance frameworks, and regulatory clarity improve.

The training gap narrows. Institutions are recognizing that deployment without training is wasted investment. Expect the 37% training figure to climb toward 60% by 2028 as structured AI training becomes a standard part of banking operations.

Cross-institution patterns emerge. Banks will converge on similar agent architectures for common use cases like compliance monitoring, fraud detection, and customer service. The differentiation will shift from “who has agents” to “who has the best agents for specific high-value workflows.”

Canada continues to lead. The structural advantages of concentrated banking, strong research institutions, and progressive regulation create favorable conditions for Canadian banks to stay at the forefront of financial AI.

The professionals who understand both the technology and the business context will be the ones building these systems. That is the real opportunity inside the agent gap. Not just the market size, but the fact that the people who bridge the space between “we have AI” and “we have agents in production” will shape how an entire industry operates for the next decade.