A new desktop AI companion just picked up serious backing. IrisGo, a startup building a proactive AI assistant for PCs, closed a $2.8 million seed round led by Andrew Ng’s AI Fund earlier this year, with Nvidia and Google also joining the cap table, according to TechCrunch AI. The company is betting that knowledge workers will pay for an agent that watches what you do once, then handles it for you forever after.
IrisGo was co-founded by Jeffrey Lai, a former Apple engineer who worked on the Chinese-language version of Siri. The name is a small wink at that history: Iris is Siri spelled backward.
What it actually does
The pitch is straightforward. Show the app how to complete a task once, and it remembers the workflow for next time. No re-prompting, no rewriting instructions.
In a demo with TechCrunch AI, Lai walked the agent through ordering a latte from Philz Coffee, including filling out credit card details and hitting purchase. He then asked IrisGo to repeat the order on its own. It did.
Coffee runs aren’t the real use case. Iris ships with a built-in skills library for things like:
- Email drafting
- Invoice processing
- Report building
- Document summarization
- A coding assistant in the vein of OpenAI’s Codex or Anthropic’s Claude Code
The app also watches your desktop behavior and quietly adds new patterns to its list of automatable tasks.
Why this matters
Proactive agents are the next frontier the industry keeps pointing at. The idea: systems that anticipate what a knowledge worker needs and just handle it, instead of waiting for a prompt. Most current AI tools still sit in a chat window and require you to describe the work. IrisGo wants to flip that, with the human doing high-level thinking while the agent grinds through clerical tasks in the background.
“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 AI. His read is that even with today’s frontier models, office work still feels manual.
What stands out here is the on-device angle. IrisGo processes a lot of data locally, which gives it a privacy story most cloud-first agents can’t tell. It’s still a hybrid architecture, so heavier tasks go to the cloud, but the company says that only happens with explicit user authorization and end-to-end encryption.
The distribution play
Credibility-by-association is doing real work here. Lai met Ng through a Carnegie Mellon alumni connection, demoed the product, and walked out with a lead investor whose name carries weight across the AI ecosystem. Ng co-founded Google Brain, so his endorsement opens doors that a cold seed pitch can’t.
The more interesting move is the OEM strategy. IrisGo just launched beta apps for macOS and Windows and is chasing deals to get preinstalled on new laptops. Acer is already signed, and Lai says more device makers are in the pipeline.
What to watch
If the preinstall strategy works, IrisGo could land in front of millions of users without spending on acquisition. That’s a sharp contrast to most AI startups burning cash on growth. The bigger question is whether “record once, replay forever” holds up when websites change, forms break, and workflows drift, which is exactly where browser-based agents have stumbled before.
For knowledge workers and IT buyers, expect a wave of similar desktop agents in the next year. The on-device privacy pitch will likely become a standard line in this category.
Full details at TechCrunch AI.