Vibe coding AI tools: the hidden costs killing mortgage builds
The pitch sounds clean: use Claude or ChatGPT to build your own internal tools, skip the vendor fees, and ship faster than your competition. Mortgage lenders are trying exactly that. The reality is messier.
Industry advisors quoted in American Banker warn that token costs, compliance exposure, and staff turnover can quickly erase whatever savings companies expected when they chose to build over buy.
The token cost surprise
AI platforms like Anthropic’s Claude and OpenAI’s ChatGPT charge for development services based on metered token consumption. Costs per task are typically not predetermined. That ambiguity is biting companies that did not model usage carefully before committing.
According to a 2025 Gartner study, more than 40% of agentic AI projects are likely to be abandoned by end of 2027. Unexpected costs and uncertain return on investment are the primary reasons cited.
Devin Zito, director of information services and corporate counsel at Assurance Financial, put the cost question in context: building with AI still costs far less than hiring a full development team, but the technical staff directing the tools need enough knowledge to get reliable output. The tools do not run themselves.
Compliance does not get easier
The ease of spinning up a custom tool does not change the regulatory requirements underneath it. Government-sponsored enterprises are pressing mortgage companies to understand their AI technology with the same rigor they apply to every other aspect of their business.
Zito, who also practices as a business and technology attorney, framed it directly: if a vendor uses AI to deliver a solution, the mortgage company still has to run due diligence on that AI. If the company builds in-house, it becomes the vendor and has to run that same due diligence on itself.
Maintenance and staff turnover
Building the tool is only the front-end cost. John Geertsema, managing principal at financial services advisory firm Capco, pointed to personnel churn as a persistent maintenance risk. When a tech executive or developer leaves, institutional knowledge of a custom-built system walks out with them.
Erik Eggers, chief revenue officer at Rocktop Technologies, put it plainly: AI and vibe coding collapsed the cost of building software on the front end, but they did nothing to collapse the cost of everything around the tools.
On the vendor side, InstaMortgage CEO Shashank Shekhar raised the opposite risk: there is no guarantee a third-party product will keep pace with AI improvements unless the contract explicitly requires it, and even then it is hard to verify.
When building still makes sense
Geertsema’s framework for the build-vs-buy decision: if the capability already exists in the market and multiple vendors offer it well, buy. If you have a specific competitive differentiator that no off-the-shelf product can replicate, that is when building earns its cost.
Token prices are also expected to fall over time, as Eggers noted. Providers are working to make models more efficient as users and vendors both get better at controlling their usage. The math on building may improve. It just does not look as simple right now as the demos make it appear.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
