The Rise of Autonomous Finance

Autonomous software agents are moving into the back office of finance, and accounts receivable is where they are landing first. As such, the task of chasing overdue invoices, historically repetitive and expensive to staff, has become an early test case for AI that acts on a company’s behalf rather than simply assisting a person.

It marks a change from software that helps finance teams do their work to software that does the work itself.

That change is already in production at Monk, a New York company building AI agents for finance teams. Full-stack software engineer Francesco Coacci, who owns the collections side of the product, offers a window into where the category is heading.

AI Moves Into the Back Office of Finance

For years, automation in finance meant dashboards and reminders that still left a person to do the actual work. The newer approach is different: autonomous AI agents are beginning to take over core financial operations to help businesses run more efficiently. Accounts receivable, which is the chasing and collecting of overdue invoices, is emerging as one of the first areas to be automated this way. A category sometimes called autonomous collections has formed around agents that pursue payment with little human involvement, drafting outreach, tracking responses and following up until an invoice clears.

Collections is a logical place to start. The work is repetitive, unglamorous and costly to staff, which makes it a natural early target for software that can run a process from end to end. It is also a sizable problem. About half of all U.S. business-to-business invoices are paid late, according to credit insurer Atradius, and the firm attributes much of the delay to administrative inefficiencies in customers’ payment processes, the kind of routine, rules-based work that agents are built to absorb.

Monk was among the first companies to apply the model, according to Coacci.

The shift fits a broader move toward agentic software that carries out tasks rather than simply assisting the people doing them. “Our collections agents perform very well on collections,” Coacci said. “We were one of the first to introduce this concept.” What began in collections has become a foothold for agents to expand across the rest of the financial back office, helping American businesses become more and more efficient, according to Coacci.

Coacci On Why It’s Happening Now

The enabler for this new method, as Coacci explains, is writing. Earlier automated notices were easy to ignore because they read as automated. Agents that produce messages hard to distinguish from a person’s draft recover far more of what is owed than the legacy, boilerplate reminders, according to Coacci. “It’s incredible what you can do if you actually send an email that sounds human,” Coacci said.

Tone is part of the design. Monk’s agent, named Julia, runs through a workflow engine that scans each conversation in real time to match the tone to the recipient’s situation. It can soften for a customer facing a cash crunch or firm up for a persistent non-payer. That control, Coacci explains, is what makes the approach workable for contact as sensitive as money owed.

The economics help explain the timing, Coacci said. Recovery rates improve with little added labor, and the cost of building such systems has fallen. Faster development cycles can help small teams ship features more quickly, and many companies are glad to hand off a function few want to run in-house. Early use by large, recognizable customers has also helped showcase this model’s potential, according to Coacci.

Together, those factors moved autonomous collections from concept to working product in a short window, Coacci said.

An Engineer’s Path to Monk

Few engineers have a closer view of where this is going than Coacci. As the full-stack software engineer who owns Monk’s collections system, he builds the part of the product that does the work.

He reached that seat by chasing hard problems. Coacci grew up in Genoa, started coding at 14 after his older brother walked him through building a video game, and was building websites for local companies by 16. He competed as a professional sailor, took a computer science degree in Italy, then moved to New York for a master’s at New York University, where he and his brother built a venture-backed marketplace for people to sell their own data and signed up about 300 users.

He joined Monk in 2025 because he wanted something harder to build, and collections gave him that. Before large language models, he recalls he used to write out a feature’s requirements and built them one at a time, shipping and iterating. Now he works closer to an architect, running several agents in parallel and steering each toward a different task while holding the overall design in his head.

“You have to make sure you know how it works under the hood, you implement, and then you look at the code of the output, not really during the iteration,” Coacci said.

For all the talk of agents taking over routine work, building them still depends on engineers who understand the problem in fine detail, and collections turned out to be exactly the kind of hard problem Francesco Coacci went looking for. That is what keeps him at Monk: the challenge of engineering an agent that does delicate work well in a high-stakes environment.