Your AI Is About to Stop Waiting for You to Ask

At Google I/O 2026 on May 19, Sundar Pichai declared the beginning of the "agentic Gemini era," where AI systems do not just respond to prompts but operate as supervised assistants capable of planning, reasoning, and taking action across products and workflows. The products Google announced are built around a genuinely different assumption about what AI is for.

Monthly token processing across Google products jumped from 9.7 trillion in May 2024 to more than 3.2 quadrillion, a sevenfold increase in one year. More than 8.5 million developers now build with Gemini models every month, while APIs process roughly 19 billion tokens per minute. That is not a company testing a theory. That is infrastructure at a scale that makes the early internet look like a pilot programme.

Google released Gemini Omni, a leap forward in world understanding, multimodality, and video editing, alongside Gemini 3.5 Flash, the first model in a new family that combines frontier intelligence with action. The word action is doing a lot of work there and it is intentional. Previous models were optimised to generate good output. This one is optimised to do something with it. The announcement that changes the most about daily life is Gemini Spark. Spark is a persistent AI agent that runs continuously on dedicated Cloud virtual machines, handling multi-step tasks in the background. Pichai described it plainly: personalized agents you set up to work in the background, 24/7, to find what you need at exactly the right moment and help you take action. Not when you open the app. Always. Antigravity, Google's agent-first development platform, has also been upgraded so developers can now build agents that help people act, not just write. For anyone building products in Nairobi right now, that compression of the tools gap is worth paying attention to.

Google DeepMind CEO Demis Hassabis said something on the sidelines that has been travelling through every tech conversation since. He said the industry is standing in the foothills of the singularity. You can debate the framing. You cannot ignore what it signals coming from someone in that seat. The shift Google announced is not about Gemini beating ChatGPT on some benchmark. It is about a fundamental change in the relationship between software and time. Software used to do things when you asked it to. The new model does things while you are in a meeting, while you are stuck on Mombasa Road wondering why traffic is moving at 4km/h again.

The products that win in the next two years will not have the best AI features. They will be the ones designed for a user who is not always present. The agent works. The human reviews. That is the new loop. If your product still requires someone to sit and type prompts to get value out of it, you are building for yesterday.