Systems of agents bring Service-as-Software to life READ MORE

A new blueprint for fintech: Goodbye human data transfer, hello AI agents as human APIs

08.01.2025 | By: Nico Stainfeld, Zach Noorani

Last year, one of our portfolio founders casually mentioned what might be the most illuminating observation we’ve heard about fintech in years. “You know what I finally figured out?” he said. “Banks don’t actually run on software. They run on humans pretending to be software.”

He’d been working late the night before, trying to understand why a simple customer verification was taking his bank three days to process. That’s when it hit him: there wasn’t some sophisticated system making decisions behind the scenes. There was a single human from compliance, manually pulling data from System A, cross-referencing it with System B, making a judgment call, then typing the results into System C.

“These people aren’t customer service representatives,” he continued. “They’re human data transfer layers.”

The metaphor was perfect. In the software world, APIs are the pipes that let different systems talk to each other seamlessly. Need to verify a login with 2FA? API call. Want to check account balance? API call. But for complex actions in traditional financial services, these actions are actually done by the human data transfer layer: people who manually take information from one system, process it (often using their own judgment and experience), and input it into another system. In other words, humans pretending to be software. Every manual handoff introduces the possibility of human error, delay, and inconsistency. And given the judgment calls often involved in these kinds of high-risk, high-regulation tasks, traditional RPA solutions have never made much headway. 

The opportunity is staggering. If we could replace this human data transfer layer with actual automation – replacing humans pretending to be software with actual software – we’d be fundamentally rewiring how financial services work.

AI Agents as “Human APIs”

This insight became even more powerful when I shared it with Tugce, founder of Eloquent, one of our newest portfolio companies. Her eyes lit up immediately. “That’s exactly what we’re building,” she said. “We’re not creating AI tools, we’re creating AI operators that can actually plug into existing backend systems and perform the same tasks humans do today.”

Eloquent’s AI operators don’t require banks or fintechs to rip out their legacy systems or spend years integrating new APIs. Instead, they learn by watching existing human workflows and then replicate those exact processes automatically, but at machine speed and scale. 

Of course, there are already dozens of AI agent startups automating everything from writing code to planning a vacation. But if a fintech or financial institution were to simply turn one of these generalist agents loose to perform highly regulated tasks, chaos would ensue, and the fines would follow. The magic is in Eloquent’s highly opinionated (not to mention highly informed) view on what the agent can, and more importantly can’t, do. These guardrails actively limit the reach of the agent so it can only work towards its immediate task without doing harm. 

We’re at the very start of a gold rush for financial-services-specific AI agents acting as human APIs. We’re betting that financial institutions will strongly prefer verticalized solutions when it comes to using AI agents. And because social proof is particularly strong among major financials, it’s clear that first movers will reap outsized market share.

Here’s how it works in practice. Take a typical fraud investigation at a major bank:

Currently, when a suspicious transaction is flagged, the customer receives an alert and reaches out to the bank. A human customer service agent – you guessed it, pretending to be software – then opens the customer’s account, reviews transaction history, checks external databases for merchant information, makes a decision, and updates multiple systems with the resolution. This process takes anywhere from 15 minutes to several hours, depending on complexity.

Eloquent’s AI operators observe this workflow, learn the decision-making patterns, and then perform the entire process automatically. They get their own usernames and passwords, log into the same systems the human would use, follow the same investigation steps, apply the same logic, and update the same databases. The difference is that it happens in seconds, not minutes or hours, with perfect consistency and documentation.

It’s hard to overstate what a sea change this represents for financial services broadly and B2B fintech infrastructure specifically. Instead of spending months or years building expensive API-to-API integrations, fintechs can now deploy AI agents to functionally build connectivity and deliver value in days, not years.

We’re seeing this pattern across our portfolio. Efficio, which helps card issuers better manage their disputes investigations workflows, is using AI agents to automatically collect documents, verify information across multiple systems, and flag known fraud patterns. Exacare is tackling senior and acute care patient admissions, where AI agents replace the complex decision-making process that traditionally requires manual review of each patient’s case file across dozens of documents. 

The transformation happening here is profound. For decades, financial institutions have been trapped in a cycle of monolithic system upgrades that cost millions and take years to implement. Most banks have given up trying to modernize their core systems because the risk and complexity are simply too high.

But AI agents that act as human APIs change everything. Instead of requiring massive backend integration projects, they work with existing systems exactly as they are. A bank can deploy meaningful AI capabilities like fraud detection, risk assessment, and customer service automation in days rather than years.

This is revolutionary for institutions with complex legacy systems. Suddenly, a bank that’s been running on 1990s or 2000s infra can have cutting-edge AI capabilities layered on top without touching their core infrastructure. The AI agents become a modern interface layer that sits on top of existing systems, breathing new life into decades-old technology.

For customers like you, this means the long-promised automation of financial services is finally becoming reality. AI agents can actually take actions like resolving disputes, updating account information without bouncing customers through multiple departments or requiring human intervention for routine tasks.

Service-as-Software, powered by systems of agents, is coming to fintech

Of course, one AI agent isn’t going to revolutionize financial services; rather, systems of many agents, working in concert, are creating a new category of intelligent middleware that can bridge any gap between systems regardless of age or complexity. 

Any given bank will ultimately have thousands or even tens of thousands of agents in production, each doing discrete tasks previously executed by the human data transfer layer. They’ll work together as a system of agents to achieve end-to-end workflows – in essence, to actually do the work required. And because the tech architecture is shifting from seat-based to outcome-based, this is going to change the business model of much of B2B fintech. We’re witnessing the emergence of an entirely new paradigm for designing, building, and pricing software.

This insight has become central to our investment thesis. When we evaluate fintech opportunities now, we’re specifically looking for companies that understand and can exploit the power of AI agents acting as human APIs. Those entrepreneurs who recognize that humans are often the weakest link in automated systems, and can replace those weak links in days not years, are building the future of financial services.

We believe this opportunity extends far beyond fintech. This goes way beyond RPA, which just doesn’t cut it for these kinds of high complexity tasks. Every industry has human data transfer layers: people manually bridging disconnected systems, making routine decisions, and transferring information between databases. Healthcare, logistics, manufacturing, legal services… they’re all running on human data transfer layers that could be replaced with AI agents acting as human APIs.

Call for startups

The next wave of B2B fintech infrastructure is about making traditional FIs and fintechs radically more efficient while providing value in days, not years. We’re investing in entrepreneurs who share this vision of a financial ecosystem where AI agents handle the routine work, freeing humans to focus on genuine relationship-building and complex problem-solving. 

The opportunity starts with identifying manual processes and building software that can replicate the human workflow exactly, without requiring expensive system integrations. Look for the places and workflows where experienced employees spend their days moving information between systems, using their judgment to make ultimately predictable decisions, and following established procedures.

We believe the era of human data transfer is ending, and the age of AI agents acting as human APIs is just beginning. If you’re building towards this future, we want to hear from you.


Published on August 1, 2025
Written by Nico Stainfeld and Zach Noorani

Related Stories