From AI Stumbles to Strategic Reset
Just two years ago, Google’s first rushed chatbot experiments damaged its credibility, with viral mistakes that cast doubt on its AI readiness. That narrative has flipped. At Google I/O 2026, the company framed a sweeping AI transformation built around Gemini and, crucially, around agentic AI—systems that can plan and act, not just chat. Gemini now counts roughly 900 million regular users, putting it on par with ChatGPT’s self-reported base, and Google is threading it through the same Search box opened by over 3 billion people daily. Behind the scenes, Google is processing 3.2 quadrillion tokens per month across its AI workloads, up from 480 trillion a year earlier. The message is clear: Gemini is no longer an add-on but the core intelligence layer powering Google’s next phase of growth and competitiveness.
Gemini 3.5 Flash: The Agentic Engine Under the Hood
At the heart of Google’s AI transformation is Gemini 3.5 Flash, positioned as its strongest agentic and coding model. Running at around twelve times the speed of comparable frontier systems, Flash is optimised for autonomous workflows instead of one-off prompts. Google highlighted internal tests where Flash independently managed research projects, executed multi-step coding pipelines, and even assembled a full operating system inside Antigravity, the company’s agentic development environment. This model now serves as the default intelligence across the Gemini app and Google Search, turning the familiar search box into a control surface for complex tasks. Rather than merely summarising the web, Flash is meant to orchestrate tools, APIs, and services on the user’s behalf, signaling a shift from static query-response interactions to dynamic, ongoing collaborations between users and AI agents.
Gemini Spark and Omni: Always-On, Any-to-Any Intelligence
Google is using agentic AI to blur the boundaries between personal assistant, operating system, and creative studio. Gemini Spark, a 24/7 personal AI agent, runs on cloud-based virtual machines so it can keep working even when users close their laptops. Tightly integrated with Gmail, Docs, Sheets, and Slides, Spark can be emailed directly and monitored through Halo on Android, effectively becoming an autonomous layer across a user’s digital life. In parallel, Gemini Omni introduces any-to-any multimodality—text, audio, images, and video in any combination—with a focus on consistent, knowledge-grounded reasoning. A demo showed Omni generating a claymation-style protein-folding explainer from a single prompt, underscoring its creative and educational potential. Together, Spark and Omni redefine Gemini from a chatbot into a pervasive, context-aware agentic AI suite spanning productivity, communication, and media.
Search, Scale, and the Business Logic of Agentic AI
Google’s most consequential move is not any single model, but the decision to make agentic AI the default behavior of Search. By wiring Gemini into the ubiquitous search box, Google can instantly expose billions of users to agentic AI products without changing their habits. This scale also strengthens Google’s economic position: advertising revenue recently rose 16 percent to USD 77 billion (approx. RM355.7 billion), helped by AI systems that deliver deeper insights into user intent. Unlike pure-play labs still wrestling with infrastructure costs, Google can amortise its massive AI data centres—fuelled by almost USD 200 billion (approx. RM924 billion) in TPU and cloud AI investment this year—across ads, consumer services, and Google Cloud. More flexible, a la carte Gemini pricing and an expanding free, ad-supported tier further widen the funnel, making agentic capabilities a default expectation rather than a premium add-on.
Beyond Assistance: The Agentic Era for Consumers and Developers
The pivot to agentic AI is reshaping Google’s relationship with both consumers and developers. For everyday users, Gemini is evolving from a helper into a semi-autonomous collaborator that can track projects, coordinate emails, and manage documents across time, not just answer questions in the moment. For developers, platforms like Antigravity and Project Genie—boosted by the integration of 280 billion Street View images into an interactive world model—offer a programmable universe where agents can test code, simulate environments, and eventually act in the physical world via connected devices. This positions Google Gemini as a direct, ecosystem-wide counterweight to ChatGPT and other rivals. Instead of competing only on raw model quality, Google is leveraging distribution, integration, and infrastructure to make agentic AI the fabric of its products, betting that the future of computing is not just intelligent software, but autonomous digital counterparts working alongside every user.