Google Launches Gemini 3.5 Flash to Run Tasks Autonomously

Google's annual developer conference on Tuesday was less about a new chatbot and more about a new category of worker. The company released Gemini 3.5 Flash, a model built for what the industry calls "agentic" work: tasks that run for hours, involve multiple steps, and require minimal human supervision. The simplest way to understand it is this. Old AI answers questions. New AI does work. You tell it to compile a competitive analysis, draft a board update from your emails and documents, or monitor a client inbox and flag anything urgent. It goes and does that, then comes back to you when it is done or when it hits something that needs your judgment. The speed aspect matters more than it might seem. Google says 3.5 Flash runs four times faster than comparable models from other providers, and they have built an even faster variant that runs twelve times faster. When an AI agent is running dozens of tasks in parallel across a business, speed directly translates to how much those tasks cost to run. Google CEO Sundar Pichai put it plainly: companies are already blowing through their annual AI budgets, and it is only May. That cost issue is real and growing. Analysts at Gartner found that agentic AI models require five to thirty times more compute per task than the older chatbot style of AI. A company that budgeted for AI last year based on a simple question-and-answer tool is now getting invoices for something far more expensive. Google is positioning 3.5 Flash as a cheaper alternative that still does frontier-level work, claiming it delivers comparable performance at less than half the price of heavier models. The personal side of the announcement is Gemini Spark, a 24/7 agent that lives in the cloud and works through your Google account. You give it a standing instruction, say monitoring your inbox for customer questions or compiling weekly status updates from your project documents, and it runs continuously whether your computer is on or not. It will roll out to paid Google AI Ultra subscribers next week. Gemini Spark enters a market that is moving quickly. Anthropic's Claude Cowork and OpenAI's ChatGPT agent are already doing comparable things. The difference with Spark is that Google already has your data. If your team works in Gmail, Google Docs, and Google Calendar, Spark connects to all of that out of the box, no configuration required. That is a real advantage over competitors who require manual setup of every integration. The limitation is the flip side of the same coin. Spark works best inside Google's world. If your organisation runs on Outlook, SharePoint, Salesforce, or any combination of non-Google tools, the out-of-the-box value drops significantly. Google says it will expand third-party integrations over time via an open standard called MCP, but those are not ready yet. For businesses already running on Google Workspace, this is worth evaluating now. The practical use cases are not futuristic: drafting status emails from project documents, flagging overdue client responses, compiling weekly digests from multiple inboxes. These are real tasks that real people spend real hours on. For businesses not in Google's ecosystem, the more important question is whether to start building agentic workflows at all, and if so, which platform to do it on. The cost issue is not trivial. One widely cited enterprise analysis found that operational costs over three years run two to three times higher than the initial build estimate, with most of the overage coming from token consumption that teams failed to model accurately upfront. Getting governance and spending controls in place before deploying agents at scale is not optional.