The promise of artificial intelligence is a transformation of everyday life and business.
However, there are practical considerations when it comes to delivering on the technology’s potential. AI models are expensive, and they don’t scale easily.
Stax Payments Chief Technology Officer Mark Sundt told PYMNTS that the models are “a mile wide and an inch deep. They answer two or three questions. It’s like, ‘Is that all there is?’ There’s a gaping hole in how these models communicate and how you get information ‘context’ between [them].”
Sundt’s comments came as part of the What’s Next in Payments series on “The Rise of Digital Labor: Exploring Agentic AI in Banking and FinTech.”
The goal is to move away from monolithic applications and limited usability (and costs incurred, step by step with each interaction) toward a distributed model that moves data across several points and improves enterprises’ workflows.
No matter the use case, fundamental questions still must be answered, including the business need and the desired outcomes of using agentic AI, he said. Any enterprise’s development teams, C-suite and timelines must all be in sync.
The move toward that distributed approach leaped forward in November when Anthropic announced the debut of the Model Context Protocol (MCP), a standard way to connect AI agents to data housed within organizations — allowing developers to build AI applications while reducing fragmentation.
Different Flavors
“There are a ton of people adopting this as it comes in a kind of ‘client favor’ where you use the models, and there’s a ‘server flavor’ where you’re exposing functionality from applications you’ve written from other large models,” he said.
The positive ripple effect is that there’s a hierarchy of models that work together to improve workflows.
MCP can (and is being used to) improve know your customer (KYC) processes — looking in real time at documentation and figuring out the “appropriate path” for data to flow and for automated decision making to take place, he said.
“I’m excited about the solutions that I’ve been able to create already using some of the libraries that are already out there,” Sundt said.
Stax was already using benefits from OpenAI’s ChatGPT Pro and other offerings; the use of MCP will only enhance those efforts to conduct deep research, he said.
In other cases, Stax used agentic AI (and MCP) in the wake of acquiring a new company in recent months, he said. The acquired firm had to undergo a PCI audit, which had to be signed off on by Sundt.
“I created a model with the PCI specifications … and I ran our findings and outcomes through the model,” he said. “I was able to identify, very quickly, that there were a few things in the compliance space that were tested but perhaps not with the rigor I had hoped.”
“It was a productive opportunity for me to vet an audit that we’ve never done before on a brand-new company that we’d just acquired,” he added.
Sundt said he used to have to write “a ton of code” for different business processes, such as vetting job applicants, but now the process is improved to the point that using AI models and MCP “with guardrails” produces better outcomes than the previously largely manual efforts.
Sundt told PYMNTS that MCP “will be the lingua franca for these models to communicate back and forth — and to ‘do’ discover and expand their capabilities on the fly. It’s absolutely game changing.”
Source: https://www.pymnts.com/
