Industry people talk about “proactive” AI all the time.
Agents that guess what you want before you even do.
Anticipate needs. Fulfill them.
No waiting.
It sounds like sci-fi, but one startup is actually building it.
Enter IrisGo.
Learning By Doing
Andrew Ng’s AI Fund led a $2.8M seed round for this company earlier this year.
He’s invested.
The premise? A desktop companion that lives on your PC.
It learns your daily workflow.
Then it automates the boring parts.
Zero prompts required.
Or at least limited ones.
Jeffrey Lai co-founded the thing.
He used to work at Apple.
Specifically? The Chinese version of Siri.
A subtle nod to his past work:
“Iris” is just “Siri” spelled backward.
Clever? Yes.
Effective? Hard to say.
But the core mechanic is straightforward.
Show the software once.
It watches.
It remembers.
It does it again.
Coffee Runs And Code
Lai demoed it for TechCrunch recently.
He wanted a coffee.
Not metaphorically.
A latte from Philz.
A popular chain in the Bay Area.
He clicked through the site.
Selected the drink.
Typed in the card info.
Clicked buy.
IrisGo recorded every step.
Then he asked the agent to do it alone.
It did.
Dutifully.
But who needs a bot to buy coffee?
Nobody really.
That’s just the tutorial.
The real target is white-collar hell.
The endless emails.
The invoice processing.
Report building.
Document summarizing.
Iris has a “skills” library.
Pre-loaded workflows for the mundane stuff.
But it also learns from your mouse movements.
Your keyboard clicks.
It spots patterns in your chaos.
There’s a lot of repetitive tasks… AI-assisted office work can still Feel incredibly manual.
That’s Lai’s pitch.
Current AI models are powerful, sure.
But talking to them? Manual.
Repetitive.
Exhausting.
The goal is autonomy.
You handle the big ideas.
The bot handles the clerical grudge work.
Privacy And Pedigree
There’s a catch.
Or maybe a feature.
Most AI eats your data in the cloud.
IrisGo processes a lot of stuff on your own device.
Local processing means tighter privacy.
Still hybrid though.
The really heavy lifting goes to the cloud.
Lai promises it’s secure.
End-to-end encryption.
Only when you explicitly say yes.
Trust is hard in AI.
So Lai borrowed some credibility.
He’s a Carnegie Mellon alum.
So is Andrew Ng.
They connected through that alumni network.
Demoed the product.
Ng liked it enough to lead the round.
Nvidia backed them too.
Google backed them.
It helps to know people in the right circles.
Does the software hold up?
Only time will tell.
But for now?
It’s just learning to order your latte.
Maybe that’s where every revolution starts.
Maybe it isn’t.
