Industry insiders say the next big thing in artificial intelligence is "proactive" systems—agents that can anticipate a user's needs and fulfill them before the user even knows what those needs are. This vision moves beyond the current generation of command-driven chatbots and virtual assistants, pushing AI into a realm where it operates with a degree of agency and prescience. Among the startups vying to lead this charge is IrisGo, a company building a desktop companion for PCs that learns about a user's daily workflows and then automates them with limited to no human prompting.
IrisGo closed a $2.8 million seed round earlier this year, led by Andrew Ng's AI Fund. The fund, launched by the renowned AI researcher and co-founder of Google Brain, has a track record of backing early-stage companies that push the boundaries of applied machine learning. In addition to the AI Fund, IrisGo has attracted investment from technology giants such as Nvidia and Google, signaling strong industry confidence in its approach.
The company was co-founded by Jeffrey Lai, a former Apple engineer who helped build the Chinese language version of Siri—the company's automated assistant. There's a subtle nod to this heritage in the name: Iris is Siri spelled backward. Lai's background gives him first-hand experience with both the potential and the limitations of voice-based digital assistants. While Siri could answer questions and set reminders, it largely functioned reactively, waiting for user commands. IrisGo aims to change that by creating an agent that observes, learns, and acts autonomously.
The core idea is elegantly simple: show the program how to do something once, and it remembers that process for future automated use. No repeat instructions are needed. During a conversation with TechCrunch, Lai ran a demo that illustrated this capability. As one watched, IrisGo recorded the steps it took to select a latte from Philz Coffee—a popular Bay Area chain—fill out credit card information, and then hit purchase. Lai then asked IrisGo to repeat the order on its own, and the agent dutifully complied.
Buying coffee, of course, is not really the point. Instead, the hope is that the system will automate a whole host of business-related tasks that currently consume huge amounts of knowledge workers' time. Iris comes with a built-in "skills" library—things like email drafting, invoice processing, report building, document summarization, and many other ready-to-use automated workflows. At the same time, Iris learns from the user's desktop behavior and automatically adds those tasks to its potential list of action items.
The application also includes a coding assistant, similar in concept to OpenAI's Codex or Anthropic's Claude Code, designed to assist developers as they go about their work. This positions IrisGo not only as a general productivity tool but also as a specialized helper for technical professionals who deal with repetitive scripting or code refactoring tasks.
"Our target audience is knowledge workers—white-collar companies. There's a lot of repetitive tasks that those workers do every day," Lai said, noting that, despite the high-octane power of today's frontier models, AI-assisted office work can still feel incredibly manual and repetitive. The goal, he said, is to move away from that and toward a more fully autonomous workflow, where the human works on high-level conceptual work while agentic systems take care of all the clerical work in the background.
A particularly appealing feature of IrisGo is that it is designed to process a lot of data on-device, giving it stronger privacy protections than other applications that rely heavily on the cloud. Lai says that the system is still a hybrid architecture—meaning that larger, more complex tasks are ultimately processed through the cloud, although the company promises that cloud processing "only occurs when explicitly authorized by the user and uses end-to-end encryption." This hybrid design strikes a balance between performance and privacy, two factors that are often in tension in AI systems.
Part of the strategy for scaling Iris has been to garner credibility through association with prominent figures and organizations. Support from Ng—notably a co-founder of the formative deep learning research team Google Brain—has helped. Lai managed to set up a meeting with Ng through a shared connection: both are alumni from Carnegie Mellon University. Lai and his co-founder demoed Iris during that meeting, and Ng's AI Fund ultimately led the startup's seed round. Nvidia and Google have also backed the company, lending further credibility and providing potential channels for integration with their own AI hardware and cloud platforms.
IrisGo recently launched the beta versions of its macOS and Windows apps. This marks a significant step toward making the product available to a broader audience beyond early testers. The company is also currently pursuing deals with laptop manufacturers to preinstall the app on new devices. It recently struck such a deal with Acer, and Lai said the hope is that the company can strike similar deals with other device makers soon. Such partnerships would give IrisGo a built-in distribution advantage, placing it directly onto the desktops of millions of users from the moment they unbox their computers.
The rise of IrisGo comes at a time when the AI industry is increasingly focused on agentic systems. Major labs like OpenAI, Google DeepMind, and Anthropic are all developing autonomous agents capable of browsing the web, manipulating software interfaces, and executing multi-step tasks. But these are often cloud-based and general-purpose. IrisGo differentiates by running primarily on the edge—on the user's own machine—and by specializing in personal workflow automation. This makes it particularly attractive to enterprises concerned about data leakage and latency.
The concept of a desktop companion is not entirely new. Early efforts like Microsoft's Clippy and Apple's Desk Accessories were rudimentary by modern standards. More recently, startups like Rewind, Mem, and Sana Labs have attempted to build AI that captures and organizes desktop activity. However, IrisGo's focus on learning through demonstration and its proactive execution of tasks sets it apart. The startup believes that the key to making AI truly useful is not just answering queries, but taking action—filing reports, sending emails, processing invoices—without waiting for a command.
Lai's background in natural language processing and on-device AI, honed during his time at Apple, gives him a unique perspective. The Chinese language version of Siri required handling tonal inflections, character-based input, and a highly diverse set of dialects. This experience taught him the value of systems that work reliably without constant internet connectivity. Parts of that philosophy are reflected in IrisGo's architecture, which is designed to function even when cloud connectivity is intermittent.
Looking ahead, IrisGo faces several challenges. Competing with incumbents like Microsoft's Co-pilot or Google's Gemini will require not only technical excellence but also rapid user acquisition. The company currently employs fewer than 20 people, but its backing by Ng, Nvidia, and Google provides both capital and credibility. If the beta attracts enough power users and converts them into paying customers, IrisGo could carve out a significant niche in the emerging category of proactive desktop AI.
Source: TechCrunch News