A few weeks ago, Peter Chapman was contemplating subscribing to software that would let his bank automate the enhanced due diligence required for some of its high-volume business customers.
"We've got a person who will pull the application, go out to the Office of Foreign Assets Control government watch lists and external data sources, write up a report and make a decision on whether to onboard the new customer," said Chapman, who is chief technology officer at Grasshopper Bank, a digital-only institution in New York.
The software he was looking at would have cost six figures annually, he said. But then, after talking it over with an internal AI working group, Chapman's team created a working model that could do this same due diligence work using AI agents in an enterprise deployment of Google Gemini called Hopper (a play on the bank's name).
"We were able to go from idea to delivery in under three weeks, and we were able to bypass going out to a vendor, which would be the traditional way of going about doing this, save ourselves money and deliver this thing internally," Chapman said.
Some would call this a sign of the "SaaSpocalypse" – the day when traditional software is no longer needed because AI agents can perform the same functions, and hundreds of software companies go out of business. The term emerged from a recent market-driven panic in which investors sold off software-as-a-service stocks due to fears that AI agents would make traditional software subscription models obsolete.
"I think there's something real about that SaaSpocalypse," said Chapman, who will be the guest on the March 17 episode of the
Raymond George, chief information officer at Clearview Federal Credit Union, isn't comfortable with the term SaaSpocalypse, but also thinks AI is going to affect software purchasing and deployment, he said during the
"I don't know if it's the end of software as a service, but I think it's going to have a big impact on it," George said. "Will it replace it or will it mutate it or just become part of how it's delivered?" Just as watching movies at home went from renting videos on physical VHS tapes or DVDs to streaming, delivery channels change all the time, he pointed out.
Theo Lau, founder of advisory firm Unconventional Ventures, takes a similar view.
"I think saying that AI will replace all software programs is overlooking the complexity of how organizations are run," she told American Banker. AI agents will be able to make certain tasks faster and easier, she said. Anthropic recently released Excel plug-ins for Claude that can help do initial financial analysis, for example.
"I don't think it can replace an entire SaaS program," because large-scale software programs do a wide variety of things. "I think it can create opportunities for solo entrepreneurs who are trying to regain time lost on manual processes."
But some finance and technology experts believe the software-as-a-service industry will dramatically change.
"Software will get liberated over time," Sumeet Chabria, founder and CEO of ThoughtLinks, a strategic advisory firm, told American Banker. "I think there's going to be a lot of creativity. How do SaaS vendors reinvent themselves versus pretending that they are going to be part of this change? That's why you see the valuation of some companies down."
Chabria, who teaches in the Chief Information and Digital Officer program at Carnegie Mellon University and was previously global chief operating officer for technology and operations at Bank of America, posted a
Unlikely to replace all software banks use
In a regulated industry like banking, companies have to demonstrate their models work correctly and that their software is audited, Lau pointed out.
"You have to actually show that and demonstrate that your modeling is fully audited and correct, there's no hallucination," she said. AI is also not good at everything, she said. For instance, it's good at gathering information, but not necessarily at writing reports.
One major example of this: It's unlikely that AI agents will replace banks' core systems, Chabria said.
"I believe AI could reverse engineer, scan and understand the guts of these systems without business documentation – a lot of the documentation around what these systems do is missing," Chabria said. "So AI could get you 90% of the way to understanding what the guts of that are and program parts of it. But AI today isn't deterministic. It's still probabilistic. So you cannot rely on it."
For banks that want to modernize dated core systems using AI, Chabria recommends the "strangler pattern," which sounds violent, but is an architectural approach for gradually migrating, modernizing, or replacing a legacy system by incrementally replacing specific functionalities with new services.
"You basically put a scaffolding around that system and you kind of repair it by removing some of the functionality away, piece by piece, and making it more hollow over time," Chabria said. "So you can do this leveraging AI, for sure. But the answer is not AI to just create a new regulatory compliant system of books and records for banking. We're not anywhere close to that."
A new breed of technology provider is cropping up to meet this need, Chabria said.
"You're going to see vendors providing frameworks and components with guardrails to build software," he said. Some startups have software that is like a car chassis, allowing the components for building a car to be used to develop new systems.
"You can build and write your own systems much more easily than buying monolithic SaaS systems and interfaces," he said. Some of these startups provide the guardrails to check compliance with banking regulations.
For bank technologists who are experimenting with creating systems using AI agents, Chabria warns that "it could end up being a huge mess if people use off-the-shelf LLMs and create software to replace production ready software, especially regulatory software." However, those who build their own infrastructure that has "regulatory scaffolding" and controls built in to create core infrastructure pieces, deploy them, and then layer agentic AI on top of that for workflow and orchestration, are more likely to succeed, he said.
"I think that's the future," Chabria said. "And I'm already seeing that happen with a couple of my clients."
