From Reactive Chatbots to Ambient AI Agents
The first wave of workplace AI arrived as chatbots: useful, but fundamentally reactive. You had to stop what you were doing, frame a prompt, then copy results back into your tools. Ambient AI agents promise a different relationship. Instead of waiting in a browser tab, a desktop AI companion lives alongside your applications, observing patterns and quietly offering or executing help. This shift is less about smarter models and more about proximity to real work. By sitting close to the operating system or productivity suite, agents can see which documents, emails, and dashboards matter at any moment. That context lets them anticipate tasks like drafting replies, structuring documents, or moving data between apps. As these agents grow more proactive, the question for professionals is no longer, “What should I ask the AI?” but “Which parts of my workflow should I let it run on its own?”
IrisGo: A Desktop AI Companion That Learns How You Actually Work
IrisGo is built around a simple observation: knowledge workers lose hours to small, repeated actions across email, spreadsheets, browsers, and internal tools. Instead of asking users to describe those workflows in prompts, IrisGo watches how tasks are completed on the desktop, then automates the steps the next time. It uses system accessibility features on Windows PCs to interact directly with apps, turning multi‑tab, multi‑click routines into background task automation. The company positions IrisGo as an on‑device ambient AI agent focused on context awareness and local learning. That means understanding which files are open, what you’re editing, and how you usually respond in similar situations. Privacy is central to its pitch: personal files, preferences, and workflow context are meant to stay on the machine by default. Backing from Andrew Ng’s AI Fund and a preloading deal with Acer signal a push to make this style of desktop AI companion feel like a built‑in part of AI‑ready PCs rather than yet another standalone app.
Gemini Spark: Google’s 24/7 Workspace Agent in the Cloud
Google is bringing similar ideas into its productivity suite with Gemini Spark, a 24/7 personal AI agent for Workspace. Instead of only answering questions, Gemini Spark can send emails, add calendar events, and complete tasks across Workspace apps, running on Gemini 3.5 and Google’s Antigravity infrastructure so it can handle long‑running jobs in the background. It also includes safeguards: the agent asks before carrying out high‑stakes actions, and users choose whether to enable it. Around Gemini Spark, Google is layering more conversational AI Workspace tools. Gmail Live lets you ask for information buried in your inbox, such as flight details, using voice. Docs Live can turn spoken rambles into structured documents, pulling in relevant context from Gmail, Drive, and the web with permission. Keep gains similar capabilities for turning speech into organized notes and lists. Together, these features push Workspace further toward ambient AI agents that manage routine coordination work while you focus elsewhere.
Productivity, Trust, and the New Division of Labor
Ambient AI agents promise a more fluid form of workplace automation, but they also raise new expectations and risks. When an assistant can see your screen, files, and communication patterns, its value depends on trust. Users need to know what is being observed, which actions are taken automatically, and when data leaves their device or cloud environment. IrisGo leans on local processing and on‑device learning to reassure users, while Google emphasizes explicit consent for higher‑impact actions in Workspace. The bigger question is how human‑AI collaboration evolves. If desktop and cloud agents reliably handle recurring tasks—summarizing files, drafting replies, scheduling meetings, preparing reports—knowledge workers may spend more time on judgment, strategy, and relationship‑driven work. But habits are hard to change. One bad automated action can undermine confidence in a tool that otherwise saves time. The next phase of AI Workspace tools will likely be defined less by raw capability and more by how gracefully these agents blend into the workday without feeling like surveillance or losing control.