Google arms enterprises with agent-building tools: Antigravity 2.0 and Gemini Spark

Most of the AI conversation over the past two years has been about generation: software that writes, drafts, summarises and suggests. At I/O 2026, Google brought a different emphasis to the enterprise: not AI that produces text for a person to act on, but agents that carry out the work themselves. The announcements Google Cloud detailed are less about a cleverer chatbot and more about the plumbing that lets a business build, run and govern autonomous agents safely. For anyone weighing where AI actually pays off inside an organisation, that shift is the headline.

What Google announced

The centrepiece for builders is Antigravity 2.0, a new standalone desktop app that Google describes as a centralised workspace to steer, customise and orchestrate agents. It is paired with an Antigravity CLI for developers who want a lighter-weight interface to rapidly build and deploy agents, tightly integrated with the desktop app. Crucially for enterprises, Antigravity is now accessible through Google Cloud’s Agent Platform, which means it inherits Google Cloud’s standard data privacy protections and Terms of Service, so, in Google’s words, agent activity runs within your secure cloud boundary by default.

Alongside the building tools, Google previewed a handful of components that round out the picture.

• Gemini Spark: a new 24/7 personal agent that works in the background across Workspace, custom connectors and the open web. You can set recurring tasks, teach it new skills and let it execute multi-step work on your behalf, while it still requires explicit approval for high-risk actions such as sending emails.
• Managed Agents API: a single API call to spin up custom agents that reason, call tools and execute code inside secure, Google-hosted remote environments, so technical teams can offload infrastructure management and focus on agent behaviour.
• CodeMender: an AI code security agent, originally developed by Google DeepMind, that autonomously identifies vulnerabilities, recommends and securely tests fixes, and can apply patches across dependent systems with your approval.

The connecting thread is the Gemini Enterprise Agent Platform. Google is positioning it as the governed foundation underneath all of this: the Managed Agents API, for instance, automatically inherits the platform’s enterprise-grade data privacy, governance and security protections rather than leaving each team to assemble its own.

Why it matters for businesses

The most interesting detail is not any single product; it is where Google has put the security boundary. Take Gemini Spark. Google says it operates in a fully managed, secure runtime on Google Cloud, with every task executing in a fresh, strictly isolated, ephemeral virtual machine so that data never overlaps between sessions. All traffic routes through a secure Agent Gateway that enforces Data Loss Prevention policies, while user credentials remain encrypted and are never exposed directly to the agent. That is a meaningful answer to the question every cautious organisation asks first: if an agent can act on its own, how do we stop it doing the wrong thing with the wrong data?

For most businesses the practical takeaway is that the infrastructure for safe, autonomous agents is maturing quickly. Isolated runtimes, explicit approval for high-risk actions, and governance inherited from the platform rather than bolted on are exactly the controls that turn an interesting demo into something a compliance team can sign off. The barrier to adoption is shifting from “can the technology do it?” to “have we scoped the right job and put the right guardrails around it?”

What to do about it

The temptation with announcements like these is to wait for the dust to settle. A better response is to start small and deliberately. Pick one multi-step workflow that is well understood and currently eats human time (reconciling records, triaging requests, preparing a recurring report) and treat it as a candidate to automate end to end. Define what a good outcome looks like before you build, decide which steps the agent may complete on its own and which require a person’s approval, and insist that it runs inside a secure, governed boundary rather than a team member’s laptop.

Done that way, a single agent earns trust on a real task and becomes the template for the next one. The platforms Google previewed are designed to support precisely this approach: build the agent, run it on managed infrastructure, and keep humans in the loop where the stakes are high.

The takeaway

I/O 2026 marks a clear move from AI that writes to agents that act. The building tools (Antigravity 2.0 and its CLI) make agents quicker to assemble; the Managed Agents API and the Gemini Enterprise Agent Platform make them safer to run; and Gemini Spark shows what an autonomous, governed agent looks like for everyday work. The opportunity for businesses is no longer hypothetical. The sensible next step is to choose a single workflow, automate it carefully, and let the results decide how much further to go.

Source: Google: Google expands Antigravity agent platform and Gemini Spark for enterprises at I/O 2026