Meet Iris: The AI That Watches and Learns
What if your computer could watch you do a task once and then just… do it for you next time? That's the core premise behind IrisGo, a new AI startup that's attracted backing from Andrew Ng — one of the most influential names in machine learning and the co-founder of Google Brain and Coursera.
IrisGo is building what its co-founders describe as an ambient desktop agent. Unlike chatbots that wait for you to ask them something, Iris runs quietly in the background, observing your workflow and gradually learning how to automate the repetitive parts of your day.
From "AI Butler" to Desktop Co-Pilot
The startup originally marketed itself under the "AI butler" label, positioning Iris as a kind of always-on personal assistant for knowledge workers. The framing has since evolved — co-founders say the butler metaphor undersold the technical depth of what Iris actually does.
At its core, Iris uses computer vision and behavioural modelling to understand what's happening on your screen. It watches you copy data between spreadsheets, fill in forms, respond to routine emails, or navigate through multi-step workflows — and builds internal models of those patterns. Over time, it surfaces automations proactively, offering to handle tasks you'd otherwise do manually.
The pitch sits somewhere between a macro recorder and a full agentic AI. It's not just replaying clicks — Iris is meant to generalize from what it observes, adapting to slight variations in context the way a capable human assistant would.
Why Andrew Ng's Backing Matters
Ng's involvement is a significant signal. He's not a spray-and-pray investor — his AI Fund typically takes concentrated positions in companies building foundational AI applications, and his endorsement carries genuine technical credibility in the industry.
The desktop AI space has gotten crowded fast. Microsoft's Copilot is baked into Windows, Apple Intelligence is coming to macOS, and a wave of startups like Rabbit and Humane have tried (with mixed results) to redefine how humans interact with computers. What differentiates IrisGo, at least on paper, is its passive learning model. You don't configure it. You don't prompt it. You just work, and it figures things out.
Privacy Questions Loom Large
Of course, an AI that watches everything on your screen raises immediate questions about privacy. Screen capture at the OS level means Iris could potentially see passwords, private messages, financial data, and confidential work documents.
The company hasn't released detailed documentation on how data is stored, processed, or protected — something it will need to address clearly before enterprise adoption becomes realistic. Early users in closed beta have reportedly been asked to sign NDAs, which hasn't helped transparency.
For consumers, the value proposition is real: time savings on grinding, repetitive computer work. The challenge will be convincing people that handing an AI continuous access to their desktop is worth the tradeoff.
What's Next for IrisGo
IrisGo is currently in limited early access. The company has not announced a public launch date or pricing. Given Ng's track record of backing startups that eventually find serious enterprise traction, it's worth keeping an eye on where Iris ends up.
The dream of a computer that learns your habits and handles the boring stuff has been around since the early days of personal computing. IrisGo is betting that AI has finally made it achievable.
Source: TechCrunch
