Evident AI Index sneak preview

You'll have to wait another week for Evident’s yearly ranking of the top 50 banks on AI. But we can offer a peek behind the Index curtain and share five eye-catching tidbits:

• 47 of the 50 banks increased their overall score this year, three jumped more than ten places and two new banks entered the top 10. 
  
  
• 47 of the 50 banks we track grew their AI headcount by at least 10% this year.
  
  
• 38 banks are now running AI training for employees, up nearly 20% from last year, and it’s increasingly being made a part of both employee onboarding and board packs. 
  
  
• 35 banks engaged with the open source community by contributing to code repositories or sponsoring open source projects, up from 30 last year.
  
  
• 32 banks we track now report AI use cases with tangible outcomes, up more than 20% from last year. 
  

STAY TUNED: The Evident AI Index and the Key Findings Report are out on Wednesday, and we’ll be talking about the results live on October 15. Sign up to attend here.

The total number of agentic use cases announced by the 50 banks we track has more than doubled in the last three months, according to brand new data from the Evident Use Case Tracker. What they’re getting out of them is less clear.

In the last 10 days alone: CommBank is using agents to prepare apps for cloud migration by analyzing code, creating documentation and rewriting them (see more on it in Use Case Corner lower down). BNY’s upgrade to Eliza lets employees use agents to draft research reports or prepare materials for client meetings. And Citi launched a pilot that pulls data and puts together research reports with a single prompt.

Real value from agentic AI, however, only comes once a bank has been “fundamentally rewired,” as JPMorganChase’s Derek Waldron put it this week. In other words, after they’ve rethought workflows and bank processes to integrate the tech and reorganized full teams around its usage. Until then, most agentic tools at banks are akin to driving a Ferrari in first gear.

Each new announcement should offer a glimpse at how banks are preparing to shift up. Not so much into what the tools actually do today, but in so far as they show the behind-the-scenes work it takes for banks to have real impact down the line with these use cases. 

The three we highlight are illustrative. CommBank’s new feature, for example, isn’t just that agents can prep applications for the cloud, it’s that they’ve built security and governance guardrails that its agents are starting to adhere to. BNY put in infrastructure robust enough to support an agentic rollout to all 50,000 employees. Citi’s achievement to us isn’t that agents can prepare research, it’s that they’re hooked into enough sources and services to do it effectively, albeit on a small scale.

For now, “there is a value gap between what the technology is capable of and the ability to fully capture that within an enterprise,” said JPMC’s Waldron. Banks still have a long way to go overhauling systems and setting up new lines of defense before that gap gets closed, and we won’t know how transformative this recent flurry of agentic use cases are until they do.

“AI machines – in quite a literal sense – appear to be saving the U.S. economy right now.”

- George Saravelos, global head of FX research at Deutsche Bank, in a client note, Sept. 23 

The Trump administration’s push for a $100,000 fee on new H-1B visa applications would, if applied to existing staff, have cost JPMorganChase nearly $250 million. It’s hardly the only bank that might feel the pain. 

The 15 American banks in the Evident AI Index have had more than 8,000 H-1B visas approved in 2025 – a vehicle they use primarily to bring on software engineers, but also some AI talent. Each bank will be faced with a choice on these roles going forward: Move jobs elsewhere, fork over the funds to keep them in the U.S. or make Trump happy and replace them with homegrown talent.

If they opt for the latter, opportunity might knock for non-U.S. banks to snap AI talent up. “Certain global experts may end up in jurisdictions outside the U.S., such as Canada," projected Tim O’Brien, managing director of North American financial institution ratings at Morningstar DBRS. 

The U.K. is already weighing making visas free for global tech talent in response to the order. And RBC’s CEO Dave McKay said he sees the fee as a “material opportunity” to bring aboard new tech talent: “From India or from South Asia or from Europe, they’ll say, ‘Well, I can’t get into the U.S., but I want to move to North America.”

Bottom line: As the Trump administration draws harder lines on immigration, bank tech talent will get caught in the middle.

Barclays elevated Antoinette O’Neill to be the bank’s new CIO, “a tremendous opportunity to accelerate our journey with AI,” she wrote on LinkedIn. O’Neill was previously COO for Barclays’ investment banking division. Craig Bright, who had previously served as Barclays’ CIO, moved to be co-COO alongside Anne Marie Darling.

HSBC hired Stephen Deakin to be its head of data platforms. He previously served as the head of architecture and shared services at Citi along with being a Citi Tech Fellow – a designation the bank reserves for those with “outstanding and sustained technical achievements and leadership.”

Santander is building an AI research team in Madrid to focus on frontier research, the head of the bank’s AI lab José Manuel de la Chica wrote on LinkedIn. 

Abu Dhabi Commercial Bank hired Shameek Kundu as Chief AI Officer. He was most recently head of financial services and chief strategy officer at TruEra and served as Standard Chartered’s chief data officer from 2014 to 2020. 

First Citizens hired Pankaj Singal as executive director, enterprise data and analytics. He was chief data officer, risk at Goldman Sachs.

BMO is looking for someone to head up machine learning and predictive analytics.

There’s a Gen AI tool that can save you time on just about anything these days. That might not actually be a great thing, new data from Stanford and BetterUp Labs contends.

“Workslop” – the ill-thought-through but well-formatted content that AI sometimes specializes in – is flooding offices, the survey found, and it threatens to eat away AI productivity gains faster than tools can generate them. Beyond just leaving workers “annoyed” and “confused,” more than 20% of the respondents said they were “offended” when they could sniff out “workslop” in their inbox.

Leaders at Goldman Sachs have been calling out overreliance on AI as a budding issue in recent weeks, with CIO Marco Argenti last week noting that it could lead to complacency. “If you have low skills and high delegation, that's not a good quadrant,” he said.

In this week’s Use Case Corner, we look at how CommBank is using agents to automate its engineering team’s grunt work and Bank of America’s new Gen AI approach to client questions.

#1 CLOUD NINE

Use Case: Lumos
Vendor: AWS
Bank: CommBank

Why it’s interesting: The bank is using agent teams to prepare legacy applications for cloud migration, getting them to assess what their specific functions are and what resources they need to plug into so the engineering team can determine how best to move them. Banks tend to have hundreds or thousands of these apps, and at CommBank, this prep work could take engineers six weeks or more. “We thought about how we [could] apply an engineering-led approach and mindset to how we migrate applications to the cloud,” said CTO Ash Moollan at an event in Sydney last month. “What would it look like to migrate this AI-first, agent-first, the smart way?”

How it works: The bank uses multiple agents to tackle different steps of the modernization process. One agent, for example, analyzes the codebase for the application and describes its function. Another updates documentation to capture how exactly it works and which systems it needs to be connected to in order to function. Another flags potential security or governance risks. Once done, an agent can even transform the code before turning it over to a human to merge. There are still humans in the loop at every step, the bank said.

By the numbers: The bank was able to migrate between 20 and 30 applications to the cloud last quarter compared to just 10 all of last year, meaning the migrations are now somewhere between eight and 12 times faster.

Use cases in similar vein: Morgan Stanley’s legacy code translation tool, PNC’s agentic system for mobile app development, NatWest’s tool to reverse engineer legacy code. 

#2 NEW DIRECTIONS

Use Case: AskGPS
Vendor: n/a
Banks: Bank of America

Why it’s interesting: An acronym for Ask Global Payments Solutions, AskGPS is a Gen AI assistant for client-facing employees that searches internal documents to give quicker answers to questions.

How it works: The bank trained the tool on more than 3,200 internal documents, including term sheets, FAQs and product guidelines. Employees can chat with the assistant and self-serve answers to questions that come in from clients.

By the numbers: Client questions could take up to an hour to answer if employees needed to phone product specialists. The bank says it expects to save “tens of thousands” of hours of employee time.

Use cases in similar vein: HSBC’s platform to answer wealth manager questions, TD Bank’s assistant to front office securities staff, JPMorganChase’s agentic assistant for investment research.

JPMorganChase’s chief analytics officer Derek Waldron gave CNBC a demo of its AI platform, LLMSuite, noting that it has roughly 125,000 daily users and can create an investment banking deck in 30 seconds. The bank has also begun deploying agentic AI across the organization, Waldron said.

TD Bank put a valuation on its AI efforts at its investor event this week. The bank says AI will help it cut out roughly $360 million ($500 million CAD) from expenses and will generate an additional $360 million ($500 million CAD) in revenue in the “medium term.” In that same timeframe, it plans to grow the share of employees “enabled with AI agents” from 20% to the full workforce.

HSBC used quantum computers to predict how likely it is that bonds would be sold at a certain price, improving their accuracy on those predictions by up to 34%, the bank revealed this week. The bank worked with IBM on the “groundbreaking world-first” application, part of its strategy to focus on “near-term application of quantum technology,” said HSBC’s quantum lead Philip Intallura. The bank was among the top banks on quantum, our report earlier this year found (see: “Banks’ quantum solace,” The Brief, March 20).

Anthropic updated its coding model, Claude Sonnet, saying that it can now run on agentic tasks for up to 30 hours. The AI lab is also demoing a generative UI – a customized landing that AI puts together based on how you use the tool.