Smaller banks are increasingly pumping the brakes on third-party vendor contracts as they realize that building AI internally has become much more feasible.
For example, Nashville, Tenn.-based Thread Bank’s AI strategy has evolved over the past several months from “basically experimenting with AI to actually engineering with it and around it,” Marty Miracle, chief digital officer and chief risk officer at the $1 billion bank, told FinAi News.
“There’s not a whole lot in the banking sector — with the way AI is matured and the right approach — that you can’t engineer for yourself, rather than be totally dependent on the ecosystem that’s around it now,” he said. “It is just far beyond what I imagined six months ago.”
Risk management tasks such as Monte Carlo simulations, which have traditionally been expensive and third-party dependent, now can be handled by internal AI tools that “take disparate risk information and collate that into one picture,” Miracle said.
Monte Carlo simulations involve mathematical models that use random sampling to better predict a wide range of financial outcomes.
“Your ability to do quantitative analysis on things was limited or expensive, and with how good AI is at things like that now, I can build my own models,” Miracle said. “I can build my own quantitative engines behind them.”
Thread Bank uses various technologies to build AI, including Anthropic‘s Claude, AI development platform Hatz AI and many types of AI sandboxes, he said.
Cutting costs
Similarly, the $1.5 billion Grasshopper Bank is becoming less reliant on third-party vendors after an aggressive push to build AI agents through Google Gemini Enterprise and other models, Chief Technology Officer Peter Chapman told FinAi News.
Grasshopper Bank has deployed dozens of in-house agents in areas including compliance, underwriting, IT and marketing, saving the digital bank significant time and money, he said.
“There’s no doubt in my mind that if we didn’t have this agentic push and this AI adoption, we would have gone out and executed agreements with six to nine new vendors over the past year, each of those costing north of $100,000 annually.”
— Peter Chapman, CTO, Grasshopper Bank
“We’re really starting to displace going out to the market with new vendors,” Chapman said.
Grasshopper is paying less than $100,000 in token costs annually for internal tools used across the organization, he said.
Vendors still needed
Banking professionals with non-technical backgrounds are turning into engineers as gen AI tools and open-source models advance, but vendors still play a crucial role in AI adoption.
Reflecting this, global funding for AI-focused fintechs reached $226.2 billion in the first quarter, a more than fivefold year-over-year increase, according to market intelligence firm CB Insights.
Thread Bank, for example, still needs help from AI vendors for compliance workflows such as anti-money laundering, Miracle said.
Grasshopper also recently launched a treasury management tool in partnership with fintech Waldo, and it works with other AI providers including EnFi, Alloy and MANTL.
Overall, banks still need vendors for some use cases, but now they are more likely to ask whether they can “build an agent for it before you go out and look to buy something off the shelf,” Chapman said.
Miracle agreed, saying “the technology’s advanced so much that we’ll be more ready to embrace doing these things ourselves as it’s expedient.”
Register here for the FinAi Lending Summit, set for Oct. 7-8 in Las Vegas. This inaugural event will include speakers from Fifth Third and Capital One as well as a fireside chat with Piermont Bank founder and Chief Executive Wendy Cai-Lee.
Facts Only
* Thread Bank's AI strategy evolved from experimenting to engineering AI internally.
* Risk management tasks like Monte Carlo simulations can be handled by internal AI tools.
* Internal AI can collate disparate risk information for Monte Carlo simulations.
* Thread Bank uses technologies including Anthropic’s Claude, Hatz AI, and AI sandboxes.
* Grasshopper Bank is building AI agents using Google Gemini Enterprise and other models.
* Grasshopper Bank has deployed in-house agents in compliance, underwriting, IT, and marketing.
* Grasshopper Bank saved time and money through agent deployment.
* Grasshopper Bank spent less than $100,000 annually on internal tools.
* Grasshopper Bank partnered with Waldo for a treasury management tool using providers like EnFi, Alloy, and MANTL.
* Global funding for AI-focused fintechs reached $226.2 billion in the first quarter.
Executive Summary
Full Take
The narrative highlights a discernible tension between the accelerating maturity of generative AI tools and the continuing necessity of external specialized services within the financial sector. The move toward internal engineering, as demonstrated by Thread Bank's focus on building quantitative engines for risk modeling, suggests a fundamental revaluation of proprietary data and analytical capabilities as a source of competitive advantage. This internal shift is not merely about cost reduction; it represents an attempt to capture the full value chain of AI development, moving from being consumers of external ecosystems to creators within them.
The contrast between the cost-saving push at Grasshopper Bank—where building agents displaced expensive vendor contracts worth over $100,000 annually—and the ongoing need for external vendors for complex regulatory compliance (like AML) reveals a segmentation in AI adoption. Banks are adopting an agentic approach for internal efficiency and quantitative analysis where the immediate benefit is clear and controllable, while still acknowledging that specialized, high-stakes functions require external expertise. The pattern suggests that vendor relationships will not vanish but will transform: they will shift from providing bespoke solutions to offering specialized, high-level services or foundational infrastructure that the in-house teams can leverage themselves.
The underlying implication concerns cognitive sovereignty: the ability of institutions to manage and direct their own AI development, rather than being constrained by external platform dependencies. If banks successfully internalize the engineering for core risk functions, it suggests a paradigm shift where complex financial modeling moves from an outsourced service to an endogenous capability, fundamentally altering the power dynamic within the industry. The remaining need for vendors underscores that while the tools change, the regulatory and implementation complexity of finance still demands specialized external knowledge. What questions remain unanswered are how this internal competency will be scaled across diverse regulatory environments, and whether building agentic systems risks creating new, opaque dependencies elsewhere in the technical stack.
Sentinel — Human
The text appears to be a human-written journalistic piece that synthesizes executive commentary from banking technology leaders to analyze the trend of banks developing internal AI capabilities.