IrisGo, a startup backed by Andrew Ng, looks to become the AI desktop buddy you never knew you needed

The promise of the seamless artificial intelligence assistant has long been a centerpiece of Silicon Valley’s roadmap, yet the reality remains largely fragmented. Users typically find themselves toggling between LLM chat interfaces and traditional productivity software. A new contender, IrisGo, backed by AI luminary Andrew Ng, is attempting to bridge this divide with a “desktop buddy” designed to observe and automate tasks directly within a user’s existing environment. Unlike the isolated web-bots of the past year, Iris is built to function as an invisible layer over the operating system, learning through observation to handle the recursive drudgery of modern office work.

This emergence comes at a pivotal moment in the evolution of Large Action Models (LAMs) and local processing. For years, automation was the domain of Robotic Process Automation (RPA)—rigid, rule-based systems that broke if a button moved three pixels to the left. The rise of sophisticated multimodal models has changed the calculus. IrisGo enters a landscape shaped by recent experiments like Rabbit’s R1 and Humane’s AI Pin, which struggled to displace the smartphone. By focusing on the desktop, where the majority of high-value professional work still occurs, IrisGo is betting that the most effective AI isn’t a new device, but a smarter inhabitant of our current ones.

At its core, IrisGo functions through a sophisticated capture-and-inference loop. The software monitors desktop activity, interpreting visual signals and metadata to understand the context of a user's workflow. This is not merely screen recording; it is the application of computer vision to identify UI elements across disparate applications—from Excel spreadsheets to proprietary CRM systems. By analyzing these patterns, the AI can theoretically replicate complex sequences, such as extracting invoice data from an email and inputting it into accounting software, without requiring the user to write a single line of script or code.

The business implications for this “agentic” approach are profound. If successful, IrisGo could disrupt the established enterprise software market by rendering specialized integration platforms obsolete. Currently, companies spend billions on Zapier or Mulesoft to get different software to talk to one another. Iris eliminates the need for these backend handshakes by interacting with the frontend exactly as a human would. This "low-friction" entry point allows the startup to penetrate industries with legacy software that lacks modern APIs, providing a competitive edge in sectors like logistics, legal services, and healthcare.

However, the path forward is fraught with significant privacy and security hurdles. An AI that "watches everything" on a desktop is a potential nightmare for corporate compliance officers and individual privacy advocates alike. IrisGo must navigate the fine line between helpful observation and intrusive surveillance. To gain enterprise trust, the startup will likely need to emphasize local processing—ensuring that sensitive data never leaves the user’s machine—and provide granular controls over what the AI is permitted to see and learn. The "black box" nature of AI learning remains a hurdle for highly regulated environments where auditability is paramount.

As IrisGo moves from its initial phase into broader deployment, the industry will be watching to see if it can solve the "hallucination" problem in action-oriented AI. While a wrong answer in a chatbot is an annoyance, a wrong action in a payroll system is a catastrophe. The next twelve months will reveal whether IrisGo can achieve the reliability necessary for professional autonomy. We are entering the era of the "agentic desktop," and the success of IrisGo will likely signal whether the next frontier of productivity lies in new applications or in a more intelligent way of using the ones we already have.