techcrunch.com web signal

IrisGo Lands $2.8M to Build Observational PC Automation Agent

andrew ng agents ai assistants ai-agents ai-assistants

Key insights

  • IrisGo learns automation workflows from a single user demonstration, removing the need for scripting or API access.
  • Andrew Ng's AI Fund led the $2.8M seed, with Nvidia and Google also participating as investors.
  • IrisGo secured a preinstall deal with Acer, targeting distribution through OEM hardware partnerships.

Why this matters

Observation-based workflow learning at the OS layer represents a direct challenge to legacy RPA vendors like UiPath and Automation Anywhere, whose enterprise contracts depend on the assumption that automation requires technical configuration. A preinstall deal with Acer means IrisGo could reach consumer and SMB users before enterprise procurement cycles even begin, compressing the typical adoption timeline for this category. Nvidia and Google's participation alongside Ng's fund suggests the investor community is treating always-on screen-aware agents as a serious infrastructure bet, not a productivity app play.

Summary

IrisGo is building a desktop automation agent that watches you work once and then replicates the workflow autonomously, targeting the gap between enterprise RPA tools and consumer AI assistants that still require manual configuration. Founded by Jeffrey Lai, a former Apple Siri engineer, the San Francisco startup raised a $2.8 million seed round led by Andrew Ng's AI Fund, with checks from Nvidia and Google also participating. The agent runs on macOS and Windows and learns from a single demonstration rather than requiring scripted workflows or API integrations. Essentially: (IrisGo, AI Fund) are betting that observation-based learning closes the setup friction that has kept desktop automation out of reach for non-technical users. - IrisGo signed a preinstall agreement with Acer and is pursuing additional OEM partnerships to distribute the agent at the hardware level. - Ng's AI Fund leading the round signals this fits a broader thesis around agentic software that operates at the OS layer rather than inside individual applications. - The $2.8M seed is small relative to the infrastructure costs of running always-on screen-capture and inference pipelines at scale. The real test is whether workflow generalization holds up across the messy, non-uniform desktop environments most users actually run.

Potential risks and opportunities

Risks

  • Always-on screen monitoring creates a high-value attack surface; a breach exposing recorded workflows or credentials captured during demonstrations could trigger regulatory scrutiny under GDPR and CCPA within months of broader OEM rollout.
  • Legacy RPA vendors (UiPath, Automation Anywhere) could accelerate their own one-shot learning features or acquire observational-AI startups to block IrisGo from enterprise accounts before it builds distribution.
  • Acer's declining PC market share means preinstall volumes may be insufficient to generate the training data and network effects IrisGo needs to compete, leaving the company dependent on a single OEM partnership that underdelivers on reach.

Opportunities

  • Enterprise security vendors (CrowdStrike, Cybereason) can position endpoint monitoring products as necessary complements to always-on desktop agents, using IrisGo's launch as a buyer-education moment.
  • Other OEMs with stronger market positions, particularly Lenovo and HP, can use a competing preinstall negotiation to extract favorable terms from IrisGo or rival agent startups in the next 90 days.
  • Workflow management and compliance platforms (ServiceNow, Zapier) could acquire or partner with observational-agent startups to add demonstration-based automation before this capability becomes table stakes across the category.

What we don't know yet

  • Whether IrisGo's always-on screen capture and local inference architecture processes data on-device or routes any workflow data through cloud infrastructure, a question with direct enterprise privacy implications.
  • Terms and exclusivity scope of the Acer preinstall deal, and whether competing OEMs (Lenovo, HP, Dell) are in active talks or contractually blocked.
  • How IrisGo handles workflow drift when underlying application UIs update, given that observation-based agents are brittle to interface changes.

Originally reported by techcrunch.com

Read the original article →

Original headline: IrisGo Raises $2.8M Led by Andrew Ng's AI Fund to Build Always-On PC Automation Agent That Learns Workflows by Observation