AI agents have spent the past year hovering somewhere between being a genuine productivity tool and conference-stage magic trick. At Google I/O 2026, Google appeared determined to move agentic AI into the first category, using its annual developer conference to announce the launch of new AI models, agents, and developer tools. Among these were the always-on personal AI agent Gemini Spark, version 2.0 of its Antigravity coding assistant, AI upgrades for search, and new subscription tiers for Gemini.
A Reuters article described the announcements as Google’s attempt to court developers and enterprise customers while competing more aggressively against OpenAI and Anthropic. Google isn’t just interested in developing and selling smarter models, but also providing the surrounding infrastructure that software companies need to build and manage AI agents as part of routine business workflows. Indeed, that same thinking has been one of the defining competitive angles of the last year, as organizations face pressure to scale agentic AI from isolated lab experiments into enterprise-wide deployment.
The launch of Gemini Spark was the most significant announcement. Rather than only responding when a user opens an app, it’s designed to run in the background on virtual machines hosted in Google Cloud. All the while, it can access Google Workspace and, over time, connect to third-party tools through the Model Context Protocol (MCP), an open standard for linking AI systems with external data and services. That’s potentially a big deal for how companies handle software work. Instead of, for example, asking an assistant to draft a message or summarize a file, a user could assign a task and have the agent keep working while the laptop is closed.
For software teams, perhaps the most important development was the launch of Antigravity 2.0, which Google said now includes a desktop app, command-line interface, and SDK for orchestrating multiple agents. This would allow developers to assign work to different AI agents and coordinate what they do and what data they can access on their chosen infrastructure.
As software teams move beyond the AI-assisted coding phase towards broader automation, the next consideration is whether and how agents can help manage broader workflows, such as drafting release notes, coordinating patches, and checking logs. The Google Developer blog shared examples of Spark monitoring product requests, recommending code changes through Antigravity, creating Jira tickets and drafting stakeholder emails.
Google also announced reduced pricing for its AI Ultra plan, along with the launch of a new $100 monthly tier aimed at developers and workplace users. CEO Sundar Pichai also said that more than 8.5 million developers currently build with Google’s models every month.
While none of this means software leaders should start rebuilding around Google’s stack, it does show that enterprise-wide agentic AI is getting closer to commercial reality. AI agents are no longer just bolt-on technologies, and they’re increasingly becoming a layer of infrastructure. That brings many promising opportunities, but also presents tougher questions about permissions, audit trails, pricing, and vendor lock-in.