Why Andrew Ng and Nvidia Are Betting Millions on This New ‘Proactive’ AI Agent

Meet IrisGo: The Andrew Ng-Backed AI Desktop Companion That Automates Your Work Before You Ask:

Backed by Andrew Ng's AI Fund, Nvidia, and Google, IrisGo is a proactive AI agent for PC and macOS that learns your workflows and automates them — with your data staying on-device.

Introduction: The Next Frontier of AI Is Proactive, Not Reactive:

The dominant AI tools most knowledge workers use today still require one thing above all else: a human to ask the question. Whether it's a chatbot, a coding assistant, or a document summarizer, the prevailing paradigm remains prompt-and-response — useful, but fundamentally reactive. Industry insiders, however, believe the next transformative shift in AI will be driven by proactive AI agents: systems that don't wait to be asked, but instead anticipate user needs and fulfill them autonomously.

IrisGo, a San Francisco-based AI startup, is building squarely toward that future. Having closed a $2.8 million seed round led by renowned AI investor Andrew Ng's AI Fund — with additional backing from Nvidia and Google — the company is developing a desktop AI companion for PC and macOS that learns a user's daily workflows and automates them with minimal to no human prompting. In a crowded field of AI productivity tools, IrisGo's vision is refreshingly concrete: show it once, and it handles the rest.

The Founding Story: From Apple's Siri to an AI-Powered Desktop Agent:

IrisGo's origin story carries a certain poetic elegance for anyone who has followed the evolution of AI assistants. The company was co-founded by Jeffrey Lai, a former Apple engineer who played a central role in building the Chinese-language version of Siri, Apple's pioneering voice-based AI assistant. The product name is a quiet nod to that heritage — Iris is, quite literally, Siri spelled backward.

The connection to Andrew Ng, one of the most respected names in deep learning and AI infrastructure, came through a shared alumni network. Both Lai and Ng are graduates of Carnegie Mellon University. Leveraging that connection, Lai secured a meeting with Ng, demoed IrisGo's capabilities alongside his co-founder, and ultimately secured Ng's AI Fund as the lead investor in the company's seed round.

The endorsement carries weight: Ng co-founded Google Brain, helped build Coursera, and has been one of the most consistently insightful voices on the practical applications of machine learning.

With Nvidia and Google also participating in the funding round, IrisGo enters the competitive AI productivity market with a rare combination: deep technical credibility, strong institutional backing, and a product narrative anchored in a clear, tangible use case — eliminating repetitive desktop work for knowledge workers.

How IrisGo Works: Learn Once, Automate Forever:

The core mechanic of IrisGo is elegantly simple, and that simplicity is a feature, not a limitation. A user performs a task on their desktop — any task — and IrisGo records and maps the workflow in the background. The next time that task needs to be completed, the agent can handle it autonomously, without any repeat instructions from the user. The technical ambition is significant: building a desktop AI agent that generalizes from observed behavior rather than relying on rigid, pre-programmed scripts.

IrisGo was shown how to place a coffee order on the Philz Coffee website — navigating the menu, selecting a latte, entering credit card details, and completing the purchase. When asked to repeat the order independently, the agent executed the entire workflow without any further input from the user. While ordering coffee may seem trivial, it serves as a proof-of-concept for far more consequential enterprise workflow automation tasks.

Beyond learned behaviors, IrisGo ships with a built-in "skills" library covering a broad range of common business tasks. These include email drafting, invoice processing, report building, document summarization, and dozens of other ready-to-use automated office workflows. The system also continuously monitors desktop activity to identify new repetitive tasks, automatically adding them to its growing list of potential automation candidates — turning passive computer use into an ever-expanding library of AI-powered productivity tools.

Built for Knowledge Workers: The Vision of Fully Autonomous Office AI:

IrisGo's target market is explicitly the knowledge economy — the vast population of white-collar professionals whose daily output is defined by high-volume, often repetitive cognitive tasks. "Our target audience is knowledge workers — white-collar companies. There's a lot of repetitive tasks that those workers do every day," Lai told TechCrunch.

The observation points to a widely shared frustration: even with access to today's most powerful frontier AI models, office work can still feel incredibly manual and fragmented. The vision Lai articulates is a meaningful evolution from the current state of AI-assisted work. Rather than a tool that a professional queries and supervises, IrisGo aims to be a fully autonomous AI workflow agent that operates in the background — handling all clerical and repetitive execution tasks while the human focuses exclusively on high-level strategic and creative thinking.

It's a compelling framing of the human-AI collaboration model: one where agentic AI handles the cognitive busywork and humans handle the judgment.

The platform also includes a dedicated coding assistant — comparable in concept to OpenAI's Codex or Anthropic's Claude Code — designed to support developers in their day-to-day workflow. This positions IrisGo not just as a productivity tool for business generalists, but as a viable AI developer assistant for technical teams, broadening its potential enterprise AI adoption footprint significantly.

Privacy-First Architecture: On-Device AI Processing as a Competitive Differentiator:

One of IrisGo's most strategically differentiated features is its approach to data privacy — a consideration that is rapidly becoming a decisive factor in enterprise software procurement. The platform is designed to process as much data as possible on-device, reducing reliance on cloud infrastructure and giving users significantly stronger privacy protections than cloud-heavy AI alternatives. In an era of growing concern about AI data privacy and enterprise data governance, this architecture addresses a real and pressing need.

IrisGo describes its underlying architecture as a hybrid model. Larger, more computationally intensive tasks are still routed to the cloud when necessary — but the company is explicit that "cloud processing only occurs when explicitly authorized by the user and uses end-to-end encryption." This opt-in, encrypted model for cloud processing represents a meaningful commitment to responsible AI development and positions IrisGo favorably against competitors whose data handling practices are less transparent.

For enterprise buyers evaluating AI productivity tools, the combination of on-device processing, user-controlled cloud access, and end-to-end encryption provides a compliance-friendly foundation. As AI agent security and data residency requirements tighten across industries, IrisGo's architecture may prove to be a long-term strategic advantage in B2B AI software sales cycles.

Growth Strategy: Hardware Partnerships, Beta Launch, and the Preinstall Play:

IrisGo's go-to-market strategy reflects a clear-eyed understanding of how consumer and enterprise software achieves scale in the PC ecosystem. The company recently launched beta versions of its macOS and Windows AI apps, opening its platform to early adopters across both major desktop operating systems. But the more structurally significant growth lever is the company's pursuit of hardware OEM partnerships — deals that would see IrisGo preinstalled on new laptops and desktop PCs at the point of manufacture.

The company has already secured its first major hardware deal with Acer, one of the world's largest PC manufacturers. Acer's decision to preinstall IrisGo on new devices provides the startup with a distribution channel that most software companies spend years — and tens of millions of dollars — trying to build. Lai has indicated that similar discussions are underway with other laptop and PC manufacturers, with the goal of making IrisGo a standard feature of the next generation of AI-powered personal computers.

The preinstall strategy mirrors the playbook that made antivirus software, browser toolbars, and more recently, AI voice assistants into ubiquitous features of the PC experience. If IrisGo can establish itself on enough new devices, it gains something far more valuable than marketing reach: behavioral data, user habit formation, and the network effects of a growing automation library — all compounding advantages in the long race to define the AI desktop companion category.

Competitive Landscape: IrisGo in the Race to Define the Agentic Desktop:

IrisGo enters a market that is attracting serious attention from some of the best-funded and most technically sophisticated companies in the world. Microsoft's Copilot is deeply embedded in the Windows ecosystem. Apple is expanding Siri's intelligence layer across macOS. OpenAI and Anthropic are both pushing toward agentic capabilities in their respective platforms. In this context, the question for IrisGo is not whether agentic AI for desktop workflows will be a large market — it clearly will — but whether a well-funded startup can carve out a defensible position before the platforms close in.

IrisGo's strongest differentiators appear to be its on-device privacy architecture, its learn-by-demonstration workflow model, and its hardware preinstall strategy. These three vectors together create a product experience that is meaningfully distinct from what AI productivity platforms from major tech incumbents currently offer. The endorsement of Andrew Ng, Nvidia, and Google — all entities with a vested interest in the broader AI agent ecosystem — also lends the company a level of technical and commercial credibility that most early-stage startups cannot claim.

The company's CMU-to-AI-Fund pipeline also hints at a talent and network advantage that could accelerate both product development and partnership formation. In the AI startup landscape of 2025, where the gap between demo and scalable product is often where companies fail, IrisGo's combination of a battle-tested founder, institutional backing, and a concrete hardware distribution deal gives it a stronger foundation than many of its contemporaries in the proactive AI agent space.

Conclusion: The AI Desktop Companion Category Is Being Built Right Now:

The era of reactive AI tools — where every output requires a human prompt — is drawing to a close. What comes next, as IrisGo and a growing cohort of proactive AI startups are demonstrating, is a computing experience where your desktop understands your work, learns your patterns, and autonomously handles your most repetitive tasks. This is not a distant vision; it is a product in beta, shipping on Acer laptops, backed by some of the most respected names in AI.

For knowledge workers drowning in clerical repetition, for developers who want an intelligent pair-programmer built into their workflow, and for enterprises looking to unlock real AI-driven productivity gains without sacrificing data privacy, IrisGo represents a compelling early answer to one of the most important questions in enterprise software: what does it look like when your computer finally starts working for you, instead of the other way around?

With Andrew Ng's conviction behind it, Nvidia and Google's resources supporting it, and Acer's distribution in front of it, IrisGo may well be one of the most important AI startups to watch in 2026 — and the desktop AI companion you didn't know you needed until it started doing your job for you.