Andrew Ng's latest investment is a startup called IrisGo - an AI agent that watches your desktop and, over time, figures out how to do your repetitive tasks without being explicitly programmed.
The company originally pitched itself as an "AI butler," a framing that captures what it's actually trying to do: observe what you're working on and handle the tedious parts before you have to ask. According to the co-founder, IrisGo learns automatically from watching user behavior on screen, rather than requiring you to set up workflows, record macros, or describe tasks in a chat box.
The concept is a more ambient version of what tools like Claudee Code](/tools/claude-code/) or Cursor do for developers - except IrisGo is targeting general desktop users. Instead of you describing a task to an AI assistant, the system builds that understanding passively by watching you work.
This is harder to pull off than it sounds. Passive observation of a desktop means the agent has to understand context across dozens of applications, identify which actions are worth automating, and avoid interfering with tasks that shouldn't be touched. The practical risk with any always-watching agent is that it automates the wrong thing at the wrong moment - or that users don't trust it enough to actually let it run.
Andrew Ng has a track record of backing focused AI automation plays through his AI Fund vehicle, including companies like Cresta and Landing AI that target specific workflows rather than general-purpose AI. IrisGo fits that pattern: narrow use case, repeated-task focus, and consumer-facing distribution.
Details on pricing, platform support (Mac, Windows, or both), and a public launch timeline haven't been disclosed. The company is still in early stages, so expect a waitlist or limited beta before any broader rollout.
For anyone who spends hours doing repetitive computer tasks - filing, reformatting, copy-pasting between apps - the pitch is genuinely interesting. Whether the execution matches the concept is the open question.