← /editions
#122

Google Gives Its Agents an Inbox

// Google Cloud's new Gemini agent runs multi-day enterprise workflows across Workspace, Microsoft 365, and Slack with multi-model support, and its persistent agents get their own Gmail, Calendar, and Drive.

$ today_in_10_seconds

Google Cloud launched a Gemini agent built to run enterprise workflows that span days, not single chat turns. It also works across Microsoft 365 and Slack and supports multiple models. The bigger story is that these agents get their own Gmail, Calendar, and Drive, which makes them something closer to staff than tools.

$ lead_story

Google Cloud's Gemini agent goes multi-day, multi-suite, multi-model

Google Cloud unveiled a general-purpose workplace agent that can research, create documents, and write code over workflows lasting several days. It reaches beyond Google's own walls into Microsoft 365 and Slack. Two design choices stand out. "Multi-day" means Google is betting on durable state, checkpointing, and resumption instead of one-shot completions. Multi-model support signals that Google expects enterprises to swap models underneath the same agent shell. For builders, the question shifts from "can an agent do this task?" to "can it hold context, permissions, and progress across a week of interruptions?"

// why_it_matters: The agent harness is becoming the product, and the model is becoming a pluggable part.

read more →

$ also_worth_attention

  • Gemini Agents get their own Gmail, Calendar, and Drive (VentureBeat)

    Google's persistent agents get dedicated Workspace storage and accounts, so they don't just borrow yours. That gives a long-running agent somewhere to keep working memory, receive replies, and schedule its own follow-ups. It also turns agent identity, auditing, and offboarding into IT problems your org now has to solve.

    link →

$ bigger_picture

The trend behind the headlines

Until now, most enterprise agents have acted as you, borrowing your OAuth tokens and your inbox. Google is moving them toward acting as themselves, with their own accounts, storage, and time horizon. That is a cleaner security model, since you can scope, log, and revoke an agent separately from a human. It also quietly makes the agent a seat-like entity, with all the licensing and governance questions that follow. If you're building agent infrastructure, identity, permissions, and durable state are where the next year of hard problems sits, more than prompting.

$ tool_of_the_day

Google Agent Development Kit (ADK)

Google's open-source framework for building multi-step, tool-using agents, and it isn't locked to Gemini models. It's a sensible place to prototype the kind of long-running, multi-agent orchestration Google is now shipping as a product. That way you can see what's under the hood before you buy the packaged version.

explore Google Agent Development Kit (ADK) →

$ subscribe --free

Get the next edition in your inbox

// DAIly Agent covers agentic AI three times a week

// free, no spam, exit anytime