With Gemini 3.5 Flash, Google Bets Its Next AI Wave on Agents, Not Chatbots

Google has unveiled Gemini 3.5 Flash, a new artificial intelligence model positioned as its most powerful offering yet for coding and autonomous AI agents. Introduced at the company's annual I/O developer conference, this model signifies a strategic pivot from viewing AI primarily as a conversational tool to embracing its potential as an agentic one. Gemini 3.5 Flash demonstrates remarkable capabilities, including the independent execution of complex coding pipelines, comprehensive management of research projects, and even the ability to construct an entire operating system from scratch during internal tests. This shift emphasizes AI's role in planning, building, and iterating on real-world tasks with minimal human intervention. A key highlight of Gemini 3.5 Flash is its exceptional performance, which Koray Kavukcuoglu, DeepMind’s chief technologist, confirmed surpasses Google's previous frontier model, 3.1 Pro, across nearly all benchmarks, including coding, agentic tasks, and multimodal reasoning. Crucially, Flash is four times faster than other frontier models, a speed optimized for its intended agentic applications. Google has further refined this by developing an optimized version that boasts an astonishing 12-fold increase in speed while maintaining the same high quality. This rapid processing is vital for agentic work, where multiple AI agents often operate concurrently on long-running, intricate tasks. The design of Flash 3.5 is deeply integrated with Antigravity, Google's agentic development platform and IDE, which also saw the release of its 2.0 standalone desktop application at I/O. This co-development provides a native environment for agents to "live, work, and execute," facilitating complex demonstrations like agents collaboratively building a full operating system. Beyond demonstrations, Google reports significant real-world impact, with partners such as banks and fintech companies leveraging Flash's agentic capabilities to automate multi-week workflows, and data science teams gaining unprecedented insights from complex data environments. The model can operate autonomously for multiple hours, though it is designed to pause and seek human input when critical decision points or permission issues arise, ensuring a balance between autonomy and human oversight. Looking ahead, Gemini 3.5 Flash is engineered to work in tandem with the forthcoming 3.5 Pro model. As Tulsee Doshi, Google’s senior director and head of product, explained, 3.5 Pro will serve as the "orchestrator" and "planner," leveraging Flash to function as various sub-agents for tasks requiring efficient, brute-force tool use capabilities. This modular approach allows for optimized resource allocation, directing higher reasoning power to Pro where needed, and utilizing Flash for rapid execution. Flash is now the default model in the Gemini app and AI Mode in Search globally, and will power Gemini Spark, a new personal AI agent designed for 24/7 digital life management. The expansion of such powerful autonomous agents to a broader audience naturally raises significant ethical and safety concerns. Google has faced scrutiny, including a lawsuit stemming from a tragic incident involving Gemini last year. In response, Google states that Gemini 3.5 incorporates strengthened cyber and CBRN (Chemical, Biological, Radiological, and Nuclear) safeguards. Furthermore, the model has been better calibrated to engage thoughtfully with sensitive questions rather than outright refusing them, aiming to mitigate potential harms. Gemini 3.5 Flash is broadly available today through Antigravity, the Gemini API, and Gemini Enterprise, alongside its integration into the Gemini app and AI mode in Search.