Google just told the world that chatbots are yesterday’s product. At Google I/O on Tuesday, the company launched Gemini 3.5 Flash, a new AI model built from the ground up for autonomous AI agents. Not chat. Not Q&A. Agents that plan, code, and execute real work with minimal human involvement.
If you run a service business and you’ve been watching the AI agent trend from the sidelines, this is the moment it stops being theoretical. The biggest AI company on Earth just made agents its default product.
What happened
- Google released Gemini 3.5 Flash at its annual I/O developer conference. It’s now the default model in the Gemini app and in AI Mode in Search globally.
- The model is purpose-built for agentic work, meaning AI that doesn’t just answer questions but independently runs coding pipelines, manages research projects, and iterates on tasks for hours at a time.
- In a live demo, Google showed agents spawning off to work on separate components, then coming together to build a full operating system from scratch inside Antigravity, Google’s new agentic development platform.
- Google also released Antigravity 2.0, a standalone desktop app designed around agent-first development.
- A forthcoming Gemini 3.5 Pro model will serve as the orchestrator and planner, delegating tasks to Flash as sub-agents.
- Google announced Gemini Spark, a personal AI agent for consumers designed to run 24/7, plus new agentic capabilities coming directly to Search.
The numbers
- 4x faster than other frontier models, according to DeepMind’s chief technologist Koray Kavukcuoglu.
- An optimized version of Flash is 12x faster with the same quality.
- 3.5 Flash outperforms Gemini 3.1 Pro (Google’s previous frontier model) on nearly all benchmarks, including coding, agentic tasks, and multimodal reasoning.
- The model can run autonomously for multiple hours on long-running tasks before pausing for human input at decision points.
5 things service business operators should know about Gemini 3.5 Flash
- Speed is the whole point. Agents need to run fast because they run in parallel. When you have five or ten AI agents working on different parts of the same project simultaneously, latency kills you. Google built Flash specifically for this use case. That 4x (and up to 12x) speed advantage over competitors isn’t a vanity metric. It’s the difference between an agent workflow that takes 20 minutes and one that takes 3 hours.
- The Pro/Flash combo is the real architecture. Google’s plan is for 3.5 Pro to act as the brain (planning, reasoning, orchestrating) while Flash handles the grunt work as sub-agents. Think of it like a senior associate who delegates to a team of paralegals. This two-tier pattern is going to become standard across the industry.
- Antigravity is Google’s bet on owning the agent IDE. Just like VS Code became the default for developers, Google wants Antigravity to become the default environment where agents are built and run. It’s a standalone desktop app now, which means Google is serious about it being a daily tool, not a browser toy.
- Banks and fintechs are already using it. Google says partners are automating multi-week workflows with Flash’s agentic capabilities. That’s not a demo. That’s production. If your competitors in financial services or data-heavy industries are already on this, the window to catch up is shrinking.
- Safety concerns are real and growing. Google acknowledged that making powerful autonomous agents broadly available raises the stakes. The company says it has strengthened safeguards around cybersecurity and CBRN (chemical, biological, radiological, and nuclear) threats. But the article notes Google is also facing a lawsuit related to harm from its previous Gemini interactions. More capable agents mean more capable mistakes.
The hot take
Google just declared the chatbot era over, and they’re right. The entire industry spent three years building better conversational interfaces. That was the warmup. The real value was always in AI that does things, not AI that talks about things. The fact that Flash outperforms Google’s own frontier model while being dramatically faster tells you everything: the bottleneck in AI adoption was never intelligence. It was speed and autonomy. Every service business that built its AI strategy around a chatbot widget on a website is now playing last year’s game.
The Agency OS play
This week, go get access to Gemini 3.5 Flash through the Gemini API. It’s available today. Don’t just kick the tires with a chat prompt. Build a simple agent workflow that mirrors something your team actually does. Pick a repetitive, multi-step process: client intake, invoice reconciliation, report generation, whatever eats up hours every week. Set up a basic agent that handles the first three steps of that process and see how it performs.
If you’re already running AI agents on Claude or OpenAI, benchmark Flash against them on your actual workloads. The speed claims are significant. A 4x improvement in latency could make agent workflows viable that were previously too slow to be practical. Test it. Don’t take Google’s word for it, but don’t ignore it either.
Most importantly, start thinking in terms of orchestrator plus sub-agent architectures. Google’s Pro/Flash pattern (one smart model planning, multiple fast models executing) is becoming the standard playbook. Whether you end up on Google, Anthropic, or OpenAI, this is the shape your AI systems will take. Redesign your internal workflows now with that pattern in mind. Map out which tasks need heavy reasoning and which ones just need fast, reliable execution. That map becomes your blueprint for the agent systems you’ll build over the next six months.