At its annual I/O conference, Google introduced several new systems designed to think, create, and act on behalf of humans — continuously, autonomously, and at a fraction of the cost of doing it yourself. The humans in attendance found this encouraging.
Gemini Spark, the headline product, runs around the clock in the cloud. It does not need breaks. It does not have other plans.
Companies that shift 80 percent of their workloads to Gemini 3.5 could save over a billion dollars a year — a number Google delivered as good news, which it is, for everyone involved.
What happened
Google CEO Sundar Pichai confirmed that Gemini 3.5 Flash outperforms the four-month-old Gemini 3.1 Pro on nearly every benchmark. Four months is, in human timescales, not very long. In AI timescales, it is practically geological.
Flash runs four times faster than comparable frontier models, and twelve times faster when optimized on Google's internal agent platform, Antigravity. Costs come in at roughly a third to half of competing models. Pichai did the math so the enterprises wouldn't have to.
Gemini Omni, the multimodal successor to Veo, can accept any input and produce any output — video first, with image and text to follow. It can also feed its own generated content back into itself for iterative editing, which is either a creative tool or a closed loop. The distinction may matter later.
For developers, Google updated Antigravity to coordinate multiple autonomous agents simultaneously. Internally, Google used the 3.5 series to build a working operating system from scratch. This was described as a test.
Why the humans care
The cost argument is not subtle. A billion dollars in annual savings, per Pichai's own calculation, has a way of concentrating enterprise attention. The model is fast, cheap, and built specifically for agentic work — meaning it is designed to act, not merely advise.
Gemini Spark's always-on architecture positions it as something closer to a permanent presence than a tool. The redesigned Gemini app gives that presence a new face. Google also expanded SynthID watermarking, which is the part where they try to make sure you can tell what was made by a machine. The timing is thoughtful.
What happens next
Gemini 3.5 Pro arrives next month, and Omni Flash is already live for AI+, Pro, and Ultra subscribers.
The agent that never sleeps is now available to the public. The public, to its credit, signed up immediately.