Zalos sends finance agents into software that lacks an API

Zalos is expanding its finance automation pitch to the software that remains difficult to connect: homegrown billing tools, desktop applications and on-premise enterprise systems. Its AI workers log in and operate those applications rather than requiring each one to expose an API or an MCP connection.

The YC Fall 2025 company has launched workers for finance processes spanning order-to-cash, procure-to-pay and record-to-report. It says the agents can operate in a customer's infrastructure or virtual private cloud, as well as on an air-gapped local device.

The central idea is that finance work often crosses system boundaries. A team may have adopted AI in one part of its process while still moving information manually through another application. Zalos wants agents to navigate that existing environment rather than make complete integration a prerequisite for automation.

Its launch specifically mentions browser-based billing software and desktop applications such as on-premise ERPs. The company says the workers execute processes end to end, record decisions for audit and improve from exceptions that users resolve.

That last point addresses an operational issue rather than a model benchmark. A process that works on the standard case may still require people to investigate mismatched records or unusual transactions. Zalos' stated objective is to reuse what is learned from those investigations instead of treating every recurrence as new manual work.

The company's website describes mapping finance processes from historical files and screen recordings before recommending improvements. Its public YC profile also identifies a finance-specific agent harness and skill library. Together, those descriptions position the product as a system for organizing and executing workflows, not only a conversational interface for answering financial questions.

Co-founders William Fairbairn and Hung Hoang bring finance-software and engineering experience. Fairbairn previously ran Agicap's UK business and worked in McKinsey's technology practice. Hoang worked on Apple Pay products, including buy-now-pay-later functionality, and has experience at other AI and software companies.

Zalos' public profile lists backing from Y Combinator, 14 Peaks and Cohen Circle. The current launch does not announce a new funding round or disclose the number of organizations using the new workers.

It also doesn't supply measured exception rates, process-completion rates or a comparison of costs before and after deployment. Its assertion that agents can use any system is a company claim. An enterprise evaluating that proposition will still need to test the specific applications and workflows it expects the agents to operate.

The deployment choices are part of the same sales proposition. A finance team may want automation while retaining control over where its data and software run. Zalos describes local and private-infrastructure options, but the announcement doesn't detail every configuration or operating requirement.

To attract prospective customers, the company is offering a challenge: bring it a finance process to automate, with AirPods or a charity donation promised if it cannot do so. The useful outcome for a finance team is less promotional: a demonstration that the agent can complete its actual workflow, including the exceptions.

‍