Andrew Ng backs IrisGo as Siri's creator builds AI desktop agent
A former Apple engineer behind Siri's Chinese-language version has raised seed funding from Andrew Ng, Nvidia and Google to build a desktop AI agent that learns workflows by watching.
The News
Jeffrey Lai, a former Apple engineer who helped build the Chinese-language version of Siri, has raised $2.8 million in seed funding for IrisGo, a desktop AI agent that learns how a knowledge worker uses their computer and then takes over the routine bits. Andrew Ng's AI Fund led the round, with Nvidia and Google joining as backers.
The product is branded simply "Iris" - a deliberate inversion of Siri. It observes a user on a Mac or Windows machine and gradually builds a library of repeatable workflows. In one demonstration, Iris watched its user order a latte from Philz Coffee once, then handled the next order alone: selecting the drink, entering card details, completing checkout. Beyond commerce, the product ships with prebuilt skills for drafting email, processing invoices, summarising documents and producing reports, plus a coding assistant pitched against OpenAI's Codex and Anthropic's Claude Code.
Lai met Ng through Carnegie Mellon alumni networks. Ng's fund led the seed after a single product demonstration. IrisGo has also signed a preinstall agreement with Acer, putting Iris on shipped laptops without the user having to download anything.
Why It Matters
The funding is modest in dollar terms but the cap table is loud. Nvidia, Google and Ng's AI Fund all writing the same cheque is the kind of signal that usually precedes a Series A within twelve months. It lands at a moment when desktop agents have become the most contested category in applied AI, with OpenAI's Operator, Anthropic's computer-use model and a wave of Chinese contenders all chasing the same surface: watch the screen, learn the task, repeat it.
What separates IrisGo from earlier agent attempts like Rabbit's R1 or the Humane Pin is the retreat from hardware. Iris is software, sandboxed on the user's existing machine, with on-device processing as the default and cloud calls only on explicit consent. That positions it for enterprise procurement teams who balked at handing employees a new gadget but will quietly add a productivity skin to laptops already in service. Lai's chosen audience: "knowledge workers - white-collar companies" who burn their days on repetitive tasks.
Indian Angle
India has roughly 5.4 million IT services employees across Tata Consultancy Services, Infosys, Wipro, HCLTech and Tech Mahindra, plus a larger captive global-capability centre workforce serving foreign banks and insurers. Most of that work - provisioning user accounts, reconciling invoices, drafting standardised client emails - is precisely the task category IrisGo is targeting. If even a modest fraction of Indian back-office workflow shifts to an Iris-style agent, the productivity gains accrue to the offshore vendors, not their Western clients, which inverts a decade of automation panic.
The Acer preinstall deal also matters here. Acer is one of the larger laptop brands in the Indian enterprise segment, and a bundled agent that ships with the machine sidesteps the IT-policy approval cycle that has slowed adoption of paid agents inside Indian banks and GCCs. RBI and SEBI rules require sensitive customer data to remain within India, so the on-device processing default is the right architecture for regulated buyers, though IrisGo will eventually have to publish a clear data-residency commitment before Indian banks sign anything in volume.
There is also a pipeline angle. Ng's AI Fund has worked closely with Indian engineers through Coursera and previously backed India-focused agent startups. An IrisGo India team, even a small one, would not surprise within the year.
FAQ
When can Indian users try IrisGo?
Beta builds for macOS and Windows are live in select markets. India availability has not been formally announced, but the Acer preinstall partnership is the most plausible route into Indian enterprises in the coming months.
How much did IrisGo actually raise?
A $2.8 million seed round, led by Andrew Ng's AI Fund, with Nvidia and Google as co-investors. The amount is modest, but the investor mix is unusually strong for a seed in 2026.
How is this different from OpenAI's Operator?
IrisGo runs primarily on-device and ships with a preset skills library rather than expecting users to script their own workflows. The retreat from cloud-by-default is its main differentiation.
What does it mean for Indian IT services firms?
It is both an opportunity and a threat. The same agents that automate offshore work for Western clients can also reduce the headcount Indian firms bill for. Pricing will need to shift from time-and-materials to outcome-based.
This story was reported by TechCrunch. Read the full original coverage at TechCrunch.