In January 2024, Twin emerged from stealth mode when AI agents were still largely conceptual. Now, the Paris-based company has partnered with Qonto, a fintech startup providing business bank accounts to over 500,000 customers across Europe, to introduce an automation agent.
Currently, there are numerous approaches for automating repetitive tasks, including API-based, no-code, or low-code automation tools like Zapier, as well as RPA software such as UiPath. However, Twin, comprising a team of nine, believes there is a more efficient method involving artificial intelligence and computer use models.
The company’s first product for Qonto, Invoice Operator, exemplifies the benefits of using artificial intelligence. With Qonto managing millions of invoices monthly, customers typically invest several hours gathering and uploading invoices. Over the past three months, Twin has developed an automatic invoice-retrieval tool to streamline this process. When using Invoice Operator, the tool retrieves a list of transactions with missing invoices and displays necessary services to access for downloading invoices alongside a browser window exhibiting the agent’s actions.
If logging into a service to download invoices is required, the browser halts, prompting users to manually enter their credentials before allowing the agent to proceed. Subsequently, Twin’s Invoice Operator autonomously locates past transactions, downloads invoices, and attaches PDFs to transactions in the Qonto account.
Hugo Mercier, Twin’s co-founder and CEO, highlighted during a product demonstration that scaling the operation to match Qonto’s requirements necessitated covering a vast array of services—reaching hundreds of thousands. This task would be infeasible with RPA, as it would require creating and continually updating custom scripts for each website. In contrast, API-based automation products, like Zapier, took a decade to support 8,000 applications. Twin, in just a few months, has already supported thousands of applications with its Invoice Operator.
Twin operates a Chromium-based web browser on a server using OpenAI’s CUA (computer-using agent) model, and was among 15 companies to trial CUA in its beta phase. This model also powers OpenAI’s Operator, a prosumer product enabling users to prompt agents for actions. Twin contends that agents capable of web browsing should provide improved performance and ease of use.
The company has focused on delivering a straightforward user experience, aiming at end-users who may not be tech-savvy. Users merely need to log into their accounts and initiate the tool, which then navigates the web to locate invoices.
Beyond invoice retrieval, Twin envisions broader applications for AI agents across various industries. Potential uses include automating order management for e-commerce, catalog classification for marketplaces, or information retrieval for call centers. Twin aspires to create a future where AI agents are more affordable, faster, and accurate in a wide array of tasks, and is exploring the possibility of developing its core agent platform, which underpins Invoice Operator, into a versatile product for developers.
