From On-Demand Tools to Ambient AI Assistants
The first wave of workplace AI lived inside chat windows: you asked a question, it returned an answer. A new generation of ambient AI assistants is flipping that script. Instead of waiting for prompts, these systems sit across your workspace, observe patterns, and start taking action on your behalf. Google’s Gemini Spark, introduced at I/O, is framed as a 24/7 AI email agent and personal operator for Workspace. It can send emails, add calendar events, and complete long-running tasks across apps while users focus elsewhere—or sleep. Meanwhile, IrisGo positions itself as an AI desktop companion that lives close to the operating system, watching how you move between email, spreadsheets, browsers, and internal tools. Together, they signal a shift from reactive chatbots to proactive workplace automation agents capable of handling routine digital chores before workers even realize they need doing.
Gemini Spark Turns Workspace Into a 24/7 AI Email and Scheduling Layer
Gemini Spark is Google’s clearest move toward autonomous workflow automation inside Workspace. Running on the Gemini 3.5 model and Google’s Antigravity infrastructure, it is designed to take actions across Gmail, Calendar, and other apps in the background. Users can delegate everyday tasks—like sending follow-up emails, booking meetings, or adding calendar events—to a 24/7 AI email agent that does not clock out. Crucially, Google emphasizes control: Spark asks before executing higher-stakes actions, and businesses can choose whether to enable it. This is complemented by new voice-driven tools—Gmail Live, Docs Live, and an upgraded Keep—that let workers speak requests and have the system search inboxes, structure documents, or turn spoken notes into organized lists. The result is a unified, conversational AI layer across the Workspace suite that reduces friction between communication, documentation, and planning, and encourages workers to offload repetitive administrative work to the agent.
IrisGo’s On-Device AI Desktop Companion Learns How You Actually Work
While Google builds into cloud apps, IrisGo embeds an AI desktop companion directly on the PC. Instead of asking users to constantly describe tasks in prompts, it observes real workflows—how you draft emails from documents, pull figures into reports, or jump between five browser tabs to complete a routine—and then automates those patterns. By using system accessibility features on Windows PCs, IrisGo can interact with multiple applications and files, turning the desktop itself into a canvas for autonomous workflow automation. Its pitch centers on context awareness and local learning: understanding computer usage, accessing local files, and automating repetitive workflows after watching them a few times. Privacy is a core differentiator. IrisGo stresses on-device processing so personal files and workflow context remain local by default, an important promise at a time when workers worry about granting broad data access. Backing from Andrew Ng’s AI Fund and a reported seed round of USD 2.8 million (approx. RM12.9 million) underscore investor confidence in this ambient AI assistant approach.
What Changes for Knowledge Workers: Delegation, Trust and Workflow Redesign
As ambient AI agents mature, the practical implications for knowledge workers go beyond saving a few clicks. Routine communication—status emails, scheduling, simple follow-ups—can increasingly be delegated to workplace automation agents like Gemini Spark, freeing people to spend more time on strategy, analysis, and relationship-building. Desktop-first tools such as IrisGo push this further by learning multi-step routines across apps. Tasks that once required careful manual navigation—summarizing local files, assembling reports, or updating internal tools—can be packaged into reusable automations. That will likely encourage teams to rethink workflows around what an ambient AI assistant can reliably handle in the background. However, trust becomes a central design challenge. Workers will expect transparency about what is being observed, what actions are taken autonomously, and when human approval is required. Early adoption patterns suggest many are ready to hand off repetitive tasks—as long as agents prove dependable, respect privacy boundaries, and make their impact on productivity tangible.