AI manifest destiny

With the use cases they’re now rolling out, top banks in the Evident AI Index for Banks appear to have a new mission: to boldly go where no AI has gone before.

Nearly two-thirds of the use cases rolled out this past quarter by our Index’s top 10 lenders broke new ground inside those firms – either the first use of AI in a new line of business or the first time AI was handling a new task. That’s up from 45% compared to Q4 last year, Evident analysis shows.

It’s not a case of these banks merely playing catch-up. These 10 lenders already account for one quarter of the 1,100 use cases lenders have rolled out publicly since 2021. Instead, their rapid push into new areas suggests the architecture they’ve built to help them reuse tech is finally helping them bring new tools into production at warp speed. Now they just need to prove the returns stack up that fast, too.

That’s closer than ever, banks insist. “There's this underappreciation for the benefits that those with scale have when it comes to AI,” Charlie Scharf, CEO of Wells Fargo (#5 in the Evident AI Index for Banks) told investors Tuesday. “We can go build an agent that can do the work in our call centers…but deploy it over a very large part of the company here.”

TD Bank (#10) is showing what that looks like in practice. Earlier this year, the Canadian lender spent eight months building a virtual assistant for its contact centers. This summer, it reused the underlying technology to roll out a similar tool for bank branch employees in just two months. And it’s not just chatbots. This week, the bank rolled out new agents for mortgage funding and discharge – getting loans paid out and wrapping up the paperwork when mortgages are paid off – that were built on the work the bank did this spring to roll out an agentic tool for the pre-adjudication process (see: “House rules,” The Brief, May 28).

“A core tenet of our AI strategy is to build once and use many times, scaling AI through repeatable patterns that lead to faster AI deployments and reduced cost of delivery,” said Raymond Chun, TD Bank’s CEO, during earnings earlier this year.

Being faster means banks can spend more time pushing AI into the parts of the business where AI use is more closely tied to saving or making the bank money. That’s happening more, as you see in the chart above. Five of the top 10 banks have rolled out their first security and IT tools so far in 2026. Some are using the tech – like Capital One (#2) is – to fix vulnerabilities across “thousands of repositories” and save potentially millions in damages from hacks (see: “The art of cyber war,” The Brief, July 23). Three have moved AI into investment banking this year.

But if banks want AI returns to come through fully, they don’t need a blockbuster tool; they need a portfolio full of them. Until that happens, the gains won’t be quite as visible as they’d like – even as the new reseases cover more ground.

“The benefits [of AI] aren't observable in our current quarterly earnings,” JPMorganChase CFO Jeremy Barnum said this week. “As important and dramatic and exciting as this theme is, it takes a long time to really do things that are visible at an institution of our size.”

That’s what JPMorganChase is – er, maybe – spending on AI annually, CFO Jeremy Barnum told investors this week, revising a previous estimate of $2 billion last year. AI is “pretty well diffused inside the whole institution,” he said, making AI costs hard to pin down. The $2 billion in benefits, however? That’s still “accurate and correct,” though it includes savings from old-fashioned ML, not just the flashy agents the bank’s been rolling out. Barnum is tired of counting, though: “I'm just very reluctant to spend a lot of energy quantifying [AI returns],” he said. “It winds up wasting a lot of people's time inside the institution.”

Between the lines: The timing here is convenient. Bank of America last month reported $800 million in annual benefits from $400 million spent on Gen AI, a twofold return (see: “Stat of the week,” The Brief, Oct. 1). JPMC’s revised figures put its returns closer – however reluctantly the bank says they were offered – to threefold. The one-upsmanship doesn’t appear over. On Bank of America’s call the day after JPMC’s, CEO Brian Moynihan said his returns were actually “two to three times” the expense.

Zoom out: Capital One has long warned against treating AI like a short-term ROI contest. “Focus on short-term ROI is why many companies never make the technology transformation that unlocks long-term ROI,” Prem Natarajan, the bank’s enterprise AI chief, told us last year. Now JPMC sounds much the same. But it’s hard to imagine that staying the case as more banks share their returns. Citi CEO Jane Fraser, for example, said this week that its AI benefits “significantly outweigh” this year’s investments. Fraser pointed to its adoption of coding tool Devin as a way to insource work and cut 30% from its tech vendor spend. If the bank came out and said its overall return from AI was fourfold, it’s hard to imagine the rest of Wall Street content to let that be the last word.

“We're not against agents or for agents, something like that. They're tools. You use them for customers.”

–Jamie Dimon, CEO at JPMorganChase, during earnings, Oct. 13

Banks are raiding tech companies to build out their AI workforce, new data from Evident shows.

Since last year, Silicon Valley firms have been the biggest feeders to bank AI teams, outside of other financial institutions. Tech companies leapfrogged consulting firms and academia this year, two of Wall Street’s traditional poaching grounds for AI talent.

Banks are increasingly looking for people with the technical know-how to turn transformation plans into actual bottom-line gains. Goldman Sachs offers the clearest view of what that looks like in practice. The bank has been “hiring a bunch of engineers from the outside to complement our already excellent team,” CFO Denis Coleman told investors this summer. It’s started using them to reinvent the way its developers work.

A year ago, the bank hired Matteo Pelati, a co-founder of AI startup LangDB, to work on agentic software development. He’s since built AgentGraph, a tool that maps the relationships between different parts of the codebase so AI agents can make changes without inadvertently breaking something else. In July, the bank added Ravi Teja, a Microsoft AI engineering veteran, as its global head of agentic AI. And Mehul Jain, an Oracle alum who oversaw the firm's developer assistant, Oracle Code Assist, joined as a managing director in August.

But Goldman isn’t confining new recruits to engineering. In July the bank hired Farhan Baluch, a former Microsoft executive who worked as a forward deployed engineer and “delivered multi-million-dollar efficiency gains,” as its head of AI for global banking and markets. In the same month, it hired Evan Kotsovinos, a VP of privacy and security at Google, to be head of asset and wealth management engineering.

Other banks are taking a similar approach to getting tech talent into both engineering and the business lines: Wells Fargo hired Faraz Shafiq from AWS to oversee AI product development and Andre Mansour to head up AI within wealth management. At JPMorganChase, former Google execs Jason Gelman and Dror Ayalon lead AI product development across the bank and in investment banking respectively.

“We don’t hire engineers to write code,” said Gill Haus, CIO of JPMC’s consumer arm, describing the shift in what banks are now looking for from tech hires. “We hire them to know what code to write."

Citi hired Paul McEwen as CTO and head of infrastructure. He was previously global head of technology services at UBS. McEwen’s role was previously held by David Griffiths, now Citi’s group head of AI.

BBVA hired David González as chief scientist and head of AI innovation. He was previously data, analytics and AI director at El Corte Inglés, one of the world’s biggest department store groups. González replaces Jon Ander Beracoechea, the bank’s former chief scientist, who departed for Santander over the summer.

Truist promoted Tim Brophy to chief data and governance officer. He’s been with the bank since 2025 as the chief data risk officer and served as head of technology and cybersecurity risk for employee platforms at JPMorganChase before that. The bank also hired Mike Frank to lead AI core engineering delivery, where he’ll be “helping turn the promise of artificial intelligence into meaningful, scalable capabilities across the enterprise,” he wrote. Frank was previously head of quality engineering, developer experience and insights at Wells Fargo.

James Patterson was appointed global head of data analytics for HSBC’s corporate and institutional banking arm. The remit is to bring together “HSBC’s data, our technology and engineering capabilities, and the potential of AI to create a genuine competitive advantage for our clients,” he wrote on LinkedIn.

Cory Lazzarato is now director of strategy, planning and value within RBC’s AI group. He was previously at Deloitte.

The AI race is getting crowded again. This week, Meta launched a pair of models, including Muse Wonder Full, a closed-source entry that independent evaluator Artificial Analysis said is faster and cheaper than Anthropic’s Fable 5.1 with nearly-equal performance on intelligence benchmarks. Cohere earlier this month released North 2, the sequel to its enterprise AI platform. And Mistral launched a new LLM it’s affectionately calling “Le Chonk.” The competition is good news for banks, at least those that know their way around model routing, the process of sending tasks to the models that can complete them in the fastest and most cost-effective way. JPMC has one “coming on,” Jamie Dimon told investors. It’s part of the bank’s strategy of “minimizing lock-in and maximizing optionality,” CFO Jeremy Barnum said.

Bank security chiefs' worst nightmare has come true: Cybersecurity firm CrowdStrike said South Korea’s Shinhan Bank was likely hacked by a single China-based actor armed with an AI agent. It’s the threat bankers have feared since Anthropic released Mythos this spring (see: “Much ado about Mythos,” The Brief, April 16). The hack wasn’t of the bank directly, but the agent found a way in via a third party, authorities said of the incident. It’s a reminder that the cybersecurity line item in next year’s budget isn’t going anywhere: Bank of America, for example, has brought on 1,000 consultants for “Mythos remediation,” CEO Brian Moynihan said this week.

CaixaBank deepened its partnership with Google Cloud, expanding Gemini Enterprise access to a wider chunk of the business. The Spanish bank will be using Google’s tech for data analytics and agentic AI as well as for hybrid infrastructure, the release said. The new deal comes on the heels of a reorganization around AI. The firm created a new AI and Data department led by Jorge Arandilla, a McKinsey veteran who joined the bank last year. So far, the team is made up of 40 specialists who are working across business and technology teams to identify new opportunities.

We can’t predict every turn AI will take in 2027. But as the dust settles from this year’s Index, we’re watching a few things closely. Last week, we looked at why AI hardware might matter even more in the coming year. This week: how banks deal with personal agents.

The bet, explained: Personal agents like Meta’s Muse and OpenAI’s dots may not be ready to trigger an agentic bank run just yet (see: “Banks vs. agents,” The Brief, Oct. 1). They can’t even move money between bank accounts. But if the buzz of the past two weeks continues and new capabilities get rolled out, banks will have to figure out how to let these agents operate safely – or figure out how to get their own products to do the agent’s job just as well.

What we’re seeing: Banks are taking different approaches. JPMorganChase is blocking outside agents that try to access accounts with a person’s login credentials, the bank said during earnings. “We’re not going to allow just any outside agent to come in,” Dimon said. That bank, however, is rolling out its own Smart Cash tool, which moves money on customers’ behalf. On the other side, Bank of America, BBVA and Wells Fargo joined the Personal Agent Protocol this past week, a new initiative led by Sierra and Meta to give agents their own credentials. The idea behind it: Rather than letting agents impersonate customers or blocking them outright, give banks a way to identify who is actually making a move.

What’s next: We’re watching how quickly banks open their doors to outside agents and the terms on which they do it. For now, they’re skeptical it’ll be a big factor in their businesses this year. “There's no question that agents will become a part of the way people interact,” Wells Fargo CEO Charlie Scharf told investors this week. “But the idea that that's going to contribute to mass movement to higher rate alternatives, I think, is completely overdone.”