With Gemini 3.5 Flash, Google bets its next AI wave on agents, not chatbots
Google has launched Gemini 3.5 Flash, an AI model designed for autonomous agents and coding, signaling a shift from conversational AI to agentic capabilities. This new model offers enhanced speed and efficiency, enabling AI agents to plan, build, and iterate on complex tasks with minimal human intervention. It will power new agentic features in Google Search and the personal AI agent Gemini Spark, with Google implementing stronger safeguards for broader consumer use.
Google has introduced Gemini 3.5 Flash, its latest AI model, which marks a significant strategic shift focusing on autonomous AI agents rather than chatbots. This model excels in coding and agentic tasks, capable of independently executing coding pipelines, managing research projects, and even building operating systems from scratch. Flash is designed to plan, build, and iterate on real-world tasks with minimal human input, moving beyond simple question-answering.
Koray Kavukcuoglu, DeepMind’s chief technologist, highlighted Flash’s superior performance, stating it outperforms the previous 3.1 Pro model across various benchmarks, including coding and multimodal reasoning. He also emphasized its speed, being four times faster than other frontier models, with an optimized version achieving 12 times greater speed while maintaining quality. This speed is crucial for agentic work where multiple AI agents collaborate on long-running tasks concurrently.
Flash’s design is deeply integrated with Antigravity, Google's agentic development platform, providing a native environment for agents to operate efficiently. Google also released Antigravity 2.0, a standalone application for agent-first development. The impact of Flash’s agentic capabilities is already evident, with partners in banking and fintech automating multi-week workflows and data science teams extracting insights from complex data.
While Gemini 3.5 Flash can run autonomously for extended periods, it is designed to request user input at critical decision points requiring human judgment or permission. The model is also engineered to work in tandem with the forthcoming 3.5 Pro model, where Pro acts as the orchestrator, leveraging Flash for sub-agent tasks.
Gemini 3.5 Flash is now the default model in Google’s Gemini app and AI Mode in Search. Google plans to integrate agentic capabilities into Search, allowing users to create and manage AI agents, and to power Gemini Spark, a 24/7 personal AI agent. Given the expanded accessibility of powerful autonomous agents, Google has strengthened safeguards for Gemini 3.5, including enhanced cyber and CBRN protections, and improved calibration for handling sensitive inquiries.
Related articles
The AI world is getting ‘loopy’
AI models are taking a significant leap forward with the adoption of "agentic loops," where AI agents continuously prompt each other to improve code and solve complex problems. This approach, though potentially resource-intensive, promises to unlock new levels of autonomous problem-solving and efficiency in AI applications.
Codex-maxxing for long-running work
Codex is increasingly being used by organizations to support long-running projects that go beyond a single prompt. This whitepaper by Jason Liu offers practical strategies for leveraging Codex as a persistent workspace, managing complex workflows and sustaining progress.
Nobel laureate John Jumper is leaving DeepMind for rival Anthropic
Nobel laureate John Jumper is departing Google DeepMind to join its competitor, Anthropic, after dedicating nearly nine years to DeepMind, where he led the AlphaFold team. Jumper, who shared a Nobel Prize for his work on AlphaFold, expressed gratitude for his time at DeepMind while looking forward to new endeavors.