When the twins came out of stealth in January 2024, AI agents were more of a theoretical concept than reality. Today, the Paris-based company is working with Qonto, a fintech startup that provides business bank accounts to more than half a million European customers, to release automation agents.
If you want to automate repetitive tasks, there are already several ways to tackle these issues. Some companies use API-based, no-code, or low-code automation products such as Zapier. Others rely on RPA software such as uipath.
With a team of nine, the Twin believes there is a much more efficient way to handle automation. As you can imagine, it included artificial intelligence and computer usage models.
Twin’s first product designed for Qonto, Twin’s invoice operators are a great example of why using artificial intelligence makes sense. Qonto processes millions of invoices per month. Customers also spend several hours a month collecting their invoices and uploading them to Qonto.
Over the past three months, Twin has created automated invoices and net-like tools that can speed up this process. When the user launches the Invoice Operator, Twin first gets a list of transactions with missing invoices. Next, you will see the list of services that you need to access to download invoices next to the browser window that shows the agent’s actions.
If you need to log in to the service to download your invoice, your browser will pause and ask you to enter your credentials manually. Doing this allows the agent to continue working by clicking the button.
The Twin Invoice Operator will then automatically find a list of past transactions, download the invoice, and attach a PDF to the transactions in your Qonto account.
“If you do that on the scale of Qonto, you basically need to cover the tail of a very long service. Thousands, hundreds of thousands, and soon everyone is using services that are used by everyone.”
“And it’s completely impossible with RPA, because you have to create a single custom script for each website, and you have to change the script every time the website changes,” he added.
For API-based automation products such as Zapier, Mercier said it took Zapier 10 years to support 8,000 applications on the platform. Twin already supports thousands of applications for invoice operators just a few months after starting work on the product.
Behind the scenes, the Twin runs a chrome-based web browser on the server. The startup uses Openai’s CUA (Computer Use Agent) model. In fact, the twins were one of 15 companies that tried CUA in beta.
CUA is also a model that enhances the operators of Openai, allowing agents to perform actions by entering prompts. In addition to improving performance, Twin believes that it should be easier to use agents that browse the web for you.
“We’ve done a lot to make the experience very simple. We’re targeting end users, maybe not tech-friendly. They don’t have to prompt or configure anything. They just log in and launch into their account and navigate to find the invoice,” Mercier said.
After searching for invoices, Twin believes there are many industries that can benefit from B2B agent applications. For example, agents can automatically manage e-commerce company orders, categorize market catalogs, and retrieve information about call center agents.
Twins are pitching a future where AI agents can be cheaper, faster and more accurate on a variety of tasks. Next, let’s see if a startup can turn the core agent platform that runs the invoice operator into a product that developers can start using in their own applications.
Source link
