When Twin came out of stealth in January 2024, AI agents were more of a theoretical concept than a reality. Today, the Paris-based company is releasing an automation agent in partnership with Qonto, the fintech startup that offers business bank accounts to more than 500,000 customers across Europe.
If you want to automate repetitive tasks, there are already several ways to tackle these problems. Some companies use API-based, no-code or low-code automation products like Zapier. Others rely on RPA software, such as UiPath.
With its team of nine people, Twin thinks there’s a much more efficient way to handle automation. As you may have guessed, it involved artificial intelligence and computer use models.
Invoice Operator, Twin’s first product designed for Qonto, is a good example of why it makes sense to use artificial intelligence. Qonto handles millions of invoices per month. And customers spend several hours per month gathering invoices and uploading them to Qonto.
Over the past three months, Twin has created an automatic invoice-retrieval tool that can speed up this process. When users launch Invoice Operator, Twin first fetches the list of transactions with missing invoices. It then shows the list of services that it needs to access to download invoices next to a browser window showing the agent’s actions.
If you need to log in to a service to download invoices, the browser pauses and asks you to enter your credentials manually. Once you do this, you can click on a button to let the agent continue its work.
After that, Twin’s Invoice Operator automatically finds your list of past transactions, downloads invoices, and attaches the PDFs to the transactions in your Qonto account.
“When you do that at the scale of Qonto, you basically need to cover a very, very long tail of services. Thousands, tens of thousands and soon hundreds of thousands of different services that everyone is using,” Twin co-founder and CEO Hugo Mercier said during a demo of the product.
“And that would be completely impossible with RPA because you would need to create a single custom script per website and then every time the website changes, you would have to modify the script,” he added.
As for API-based automation products like Zapier, Mercier said that it took Zapier 10 years to support 8,000 applications on its platform. Twin already supports thousands of applications for its Invoice Operator just a couple of months after starting work on the product.
Behind the scenes, Twin runs a Chromium-based web browser on a server. The startup uses OpenAI’s CUA (computer-using agent) model. In fact, Twin was one of the 15 companies that got to try CUA in beta.
CUA is also the model that powers OpenAI’s Operator, its prosumer product that lets you enter a prompt to let an agent perform an action for you. In addition to better performance, Twin believes it should be easier to use agents that browse the web for you.
“We worked a lot on making the experience extremely simple. We are really targeting the end user, maybe people who are not tech-friendly. They don’t have to prompt or configure anything. You just log into your accounts, launch it, and it navigates to find the invoices,” Mercier said.
After invoice retrieval, Twin thinks there are many industries that could benefit from B2B agentic applications. For example, agents could automatically manage orders for an e-commerce company, classify the catalog of a marketplace, or retrieve information for call center agents.
Twin is pitching a future in which AI agents become cheaper, faster, and more accurate across a wide variety of tasks. Now let’s see if the startup can turn the core agent platform that powers Invoice Operator into a product that developers can start using in their own applications.