When Twin He left Stealth in January 2024. AI agents were a more theoretical concept than reality. Today, the company based in Paris releases an automation agent in cooperation SystemStartup FinTech, which offers business bank accounts over 500,000 customers throughout Europe.
If you should automate repetitive tasks, there are several ways to resolve these problems. Some corporations use API -based interfaces, without code or low code, resembling automation Dump. Others rely on South African software, for example Uipath.
With a team of nine people, Twin believes that there is a much more efficient way of operating automation. As you possibly can guess, it included artificial intelligence and models of using computers.
The invoice operator, the first Twin product designed for Qonto, is a good example of why it makes sense to make use of artificial intelligence. Qonto serves hundreds of thousands of invoices per 30 days. Customers spend several hours a month collecting invoices and sending them to Qonto.
Over the past three months, Twin has created an automatic invoice survival tool that may speed up this process. When users run the invoice operator, Twin first downloads a list of transactions with missing invoices. Then he shows a list of services with which he must access invoices next to the browser window showing the agent’s actions.
If you should log in to the service to download invoices, the browser stops and asks you to manually introduce your certificates. After doing this, you possibly can click the button so that the agent can proceed working.
Then the Twin invoice operator robotically finds a list of previous transactions, downloads invoices and attaches PDF to the transaction on the Qonto account.
“When you do this on a Qonto scale, basically you have to cover a very, very long tail of services. Thousands, tens of thousands and soon hundreds of thousands of different services that everyone uses,” said Twin Mercier and general director during the product demo.
“And this would be completely impossible in the case of South Africa, because you need to create one non -standard script to the website, and each time the site changes, you need to modify the script” – he added.
When it involves automation products based on APIs, resembling Zapier, Mercier said that the service of 8,000 applications took 10,000 applications. Twin already supports 1000’s of applications for an invoice operator just a few months after starting work on the product.
Behind the scenes, Twin launches a chrome web browser on the server. The startup uses the CUA model (computer use agent) OPENAI. In fact, Twin was one of 15 corporations that might try CUA in beta.
CUA is also a model that drives the OPENAI operator, its prosumer product, which means that you can introduce a prompt for permission to be performed to the agent for you. In addition to higher performance, Twin thinks that it is easier to make use of agents who are looking at the Internet for you.
“We worked a lot about making this experience extremely.
After downloading the invoice, Twin believes that there are many industries that may gain advantage from B2B agency applications. For example, agents can robotically manage orders for e-commerce, classify the market catalog or download information for call center agents.
Twin throws a future in which AI agents develop into cheaper, faster and more accurate in many different tasks. Let’s see now whether running can transform the Agent Core platform, which powers the invoice operator into a product whose developers can start using their very own applications.