Google I/O 2026 put two Gemini directions in the spotlight: creative generation and task execution. Together, the nine demos of Gemini Omni and Gemini 3.5 show how Google is applying AI to video creation, coding, Search experiences, personal agents, and multi-step digital work.
Takeaways 🚀
✅Nine demos, two Gemini directions
✅Omni blends reasoning with creation
✅Gemini 3.5 powers agentic workflows
✅Search becomes more task-oriented
✅AI tools aid productivity
At the heart of it all, the core distinction is simple. With Gemini Omni, Gemini’s ability to reason meets the ability to create — most visibly through conversational video editing. Gemini 3.5, including Gemini 3.5 Flash, is built to help you execute complex, agentic workflows. In other words, think of agentic workflows as tasks where an AI system plans, takes steps, uses tools, and works toward a goal — all under your direction.
With that in mind, this article walks through all nine demos in plain language. It separates the Gemini Omni examples from the Gemini 3.5 Flash ones, explains what each demo shows, and covers why these features may matter to creators, developers, Search users, and Gemini app users.
Gemini Omni: Where Reasoning Meets Creative Video Generation
Gemini Omni generates and edits video using text, images, audio, and existing footage. Across all its demos, one central idea stands out: you describe the change you want, and the model applies it while keeping the scene coherent.
The key shift here is the interface. Rather than relying on traditional editing tools for every adjustment, you work through conversation instead. That does not remove the need for human judgment — but it does fundamentally change how you can direct a creative process.
1. How Does Gemini Omni Support Conversational Video Editing?
The first demo shows Gemini Omni editing a video from a plain-language instruction. Specifically, the user asks the model to turn a sculpture into one made of bubbles.
This goes well beyond applying a filter. First, the model identifies the object. Then it understands the material change requested and creates a new version of the scene that still connects to the original. Throughout this process, it also maintains motion, lighting, and the shot’s overall structure.
Ultimately, this demo captures what makes Gemini Omni different. The model reasons about what a bubble sculpture should look like, then creates an edited video that reflects that reasoning.
Mini takeaway: Gemini Omni lets you make detailed visual changes just by describing what you want.
2. How Can Gemini Omni Reimagine the Action in a Video?
The second demo goes even further than changing how an object looks. Instead, it shows Gemini Omni altering the action and structure of an entire scene.
In this case, the user asks for several changes at once: dimmed lights, a floating glass sphere, a black-and-white checkerboard room inside it, a recursive image effect, and a looping camera movement. That kind of instruction asks the model to manage objects, lighting, space, motion, and visual continuity — all at the same time.
What makes this demo stand out is that Gemini Omni is not just tweaking details. Rather, it reinterprets the scene based on creative direction. It understands the original video and adds new visual elements that fit the prompt.
Mini takeaway: This demo shows Gemini Omni as a tool for reworking what happens inside a video — not just changing its look or color.
3. How Does Multi-Turn Refinement Work in Gemini Omni?
The third demo introduces a step-by-step editing process called multi-turn refinement. As the name suggests, each new instruction builds on the one before it.
The process starts with a video of a violinist. From there, the user asks the model to move the violinist into a different visual environment. Next, they ask it to make the violin invisible. Finally, they request a new camera angle over the violinist’s shoulder.
This matters because creative work is almost always iterative. You rarely know the final version of a video from the start. Instead, you try a change, review it, and adjust. As a result, a good AI video tool needs to remember earlier steps and keep the scene consistent throughout.
Thanks to this approach, Gemini Omni preserves context across several edits. It follows a creative thread rather than treating each prompt as a new, unrelated request.
Mini takeaway: Multi-turn refinement lets you shape a video step by step through conversation — without losing what came before.
Gemini 3.5 Flash: Built for Agentic Workflows
While Gemini Omni focuses on video creation, Gemini 3.5 Flash takes a different direction — it focuses on action. Google positions Gemini 3.5 as a model family that combines intelligence with execution.
Before going further, a quick definition helps. An agentic workflow is a process where an AI system carries out a series of steps toward a goal. It can organize information, write code, use tools, produce outputs, and adjust as it goes. You set the direction — and the system does more than just answer a single question.
With that context in place, the Gemini 3.5 Flash demos bring this idea to life through Antigravity, AI Studio, Search, and the Gemini app.
4. How Does Gemini 3.5 Flash Handle Agentic Tasks at Scale?
One demo shows Gemini 3.5 Flash working with Antigravity to rename and categorize unstructured assets — files, media items, or other materials that need sorting based on specific criteria.
This is a practical example of agentic work. Rather than just advising you on how to organize files, the system helps carry out the process. It inspects items, applies rules, makes decisions within the task, and repeats the workflow across many assets.
Beyond that, the demo also introduces subagents — smaller AI agents assigned to parts of a larger job. In a complex workflow, several subagents can work together to complete the task more efficiently.
For businesses and teams, this kind of capability applies to file management, coding projects, content operations, and any other work that involves many repeated steps.
Mini takeaway: Gemini 3.5 Flash paired with Antigravity is built to execute multi-step workflows at scale.
5. How Does Gemini 3.5 Flash Create Interactive Web UIs in AI Studio?
Another demo shifts the focus to AI Studio, where Gemini 3.5 Flash generates user interface ideas. The example centers on a checkout flow — the screens a person moves through when buying something online.
Building a checkout flow typically involves design decisions, user experience planning, and code. In the demo, however, Gemini 3.5 Flash quickly generates different interface options for developers and designers to review.
Of course, that does not mean the first output is ready to ship. Human review still matters — especially for usability, accessibility, security, and brand needs. Even so, the demo shows how the model can significantly speed up early-stage exploration.
For instance, a product team could compare several layout ideas before deciding which one to refine. Similarly, a developer could use the output as a starting point for interactive web graphics or app screens.
Mini takeaway: In AI Studio, Gemini 3.5 Flash helps generate interface concepts and code-based experiences faster — while keeping final decisions in your hands.
Gemini 3.5 Flash in Search and Personal AI Experiences
Beyond developer tools, several demos show Gemini 3.5 Flash in action outside specialist environments. These examples focus on Search, personal agents, and custom tools that support everyday tasks.
This matters because Gemini 3.5 is not only for coders. In fact, it also powers more dynamic experiences for people who never write a line of code.
6. How Do Information Agents Work in Search?
One demo highlights information agents inside Search. Put simply, these agents monitor a topic and send updates when relevant information appears.
In the example, the user wants to know when favorite athletes announce sneaker collaborations or signature product drops. Rather than searching again and again, an agent watches for updates and summarizes them when something changes.
This is a clear, everyday example of an agentic workflow. You set a goal. The agent monitors information over time. Then, when it finds something relevant, it sends an update that helps you act.
In practice, this kind of feature works well for topics that change often — product launches, travel deals, job postings, events, or research updates. The value, therefore, lies in ongoing assistance — not just a one-time answer.
Mini takeaway: Information agents help Search move from answering single questions to supporting ongoing goals.
7. How Can Search Generate Dynamic Visual Responses?
Another demo takes a different approach, showing Search creating an interactive visual explanation of gyroid patterns. A gyroid is a complex three-dimensional structure often discussed in mathematics, science, and design.
For most users, a visual explanation works far better than text alone. With that in mind, Gemini 3.5 Flash helps Search produce a custom generative UI — an interface built specifically for that question.
This points to a broader shift in how Search may present information. Instead of only showing links, text summaries, or static images, Search can now generate a response format that fits the topic. For a visual or technical subject, that might mean an interactive diagram, simulation, or tool.
That said, you should still treat these as helpful explanations — not final authorities. For important decisions, always verify key details from trusted sources.
Mini takeaway: Gemini 3.5 Flash helps Search create interactive responses when a visual format explains a topic better than plain text.
8. How Can Search Build Custom Dashboards, Trackers, and Mini Apps?
Building on the idea of personalized experiences, a related demo shows Search creating custom tools — dashboards, trackers, and mini apps. The example in this case is a fitness tracker.
This demo shows how agentic coding becomes genuinely useful for everyday planning. You may want to track workouts, rest days, goals, or progress — but not know how to build a tool from scratch. With Gemini 3.5 Flash, Search can create a custom experience based on your request.
Moreover, the same idea applies to many other ongoing tasks:
- Planning a wedding
- Managing a study schedule
- Tracking a household budget
- Organizing a move
- Building a meal plan
- Monitoring personal habits
The practical benefit, in each case, is personalization. A general app may not fit your specific needs. A custom tracker built around your goal, on the other hand, can.
Mini takeaway: This demo shows how Gemini 3.5 Flash can turn a user request into a practical, reusable tool.
9. How Does Gemini Spark Act as a Personal AI Agent?
The final demo introduces Gemini Spark — a personal AI agent that runs on Gemini 3.5 and uses the Antigravity harness. It works alongside tools like Gmail, Docs, Slides, and other services.
In the demo, Gemini Spark creates a list of nut-free snacks and adds them to Instacart. This shows a complete task flow: understand the request, generate a suitable list, and then take action inside another app.
For everyday users, this is the clearest example of an AI agent at work. It does not just give you a suggestion — it helps you complete the task from start to finish.
This type of assistant could, over time, support routine planning, shopping, scheduling, document preparation, or organizing information across apps. Even so, personal agents also require clear permissions, user control, and trust. Before using one, you need to understand what the agent can access and what actions it can take.
Mini takeaway: Gemini Spark shows how Gemini 3.5 can help you complete multi-step tasks across your digital tools.
What Do These Nine Demos Tell Us?
Taken together, the nine demos reveal two related but distinct uses of AI.
Gemini Omni focuses on creative transformation. Its demos cover conversational video editing, reimagined video action, and step-by-step creative refinement. In each case, the common thread is the same: you guide visual creation through natural language.
Gemini 3.5 Flash, on the other hand, focuses on execution. Its demos cover file organization, UI generation, information agents, dynamic Search responses, custom trackers, and personal agents. Here, the common thread is also consistent: you ask the model to help complete a sequence of actions.
To summarize it simply:
- Gemini Omni: reasoning + creation, especially for video
- Gemini 3.5 Flash: reasoning + action, especially for agentic workflows
- Antigravity: the system that helps coordinate agents and complex tasks
- Search, AI Studio, and the Gemini app: where these capabilities reach different types of users
Conclusion: Creating vs. Doing
Overall, the nine demos of Gemini Omni and Gemini 3.5 in action show Google applying its models to two broad areas: making content and completing tasks.
On one side, Gemini Omni targets creative video work. You edit, transform, and refine scenes through conversation. On the other side, Gemini 3.5 Flash targets action-oriented workflows. It helps you organize assets, generate interfaces, monitor information, build tools, and support personal tasks.
For general users, the most important takeaway is not that everything becomes automatic. Human direction, review, and judgment still matter. What the demos do show, however, is a new set of ways to work with AI — describe the video you want to change, define the task you want done, or ask for a tool built around your goal.
Ultimately, if your work involves visual creation, Gemini Omni is the model to watch. If, instead, your work involves repeated steps, coding, Search tasks, or personal productivity, then Gemini 3.5 Flash and agentic workflows deserve your attention.
