Google I/O 2026: The Era of AI Agents

Gemini Omni — The centre of everything

This is the most anticipated product, and also the one that made the strongest impression at I/O 2026.

Gemini Omni is a new generation of multimodal models developed by Google DeepMind, built on an "Any-to-Any" philosophy — accepting input from any format (text, image, audio, video) and generating corresponding output. The first version, Gemini Omni Flash, focuses on video generation and editing.

What makes Omni different?

Unlike previous AI video tools that only accepted text commands, Omni allows users to engage in multi-turn conversations to edit videos step by step, with each edit changing only the requested element without disrupting the rest. For example:

"Change the background to a sunset."
→ "Add a character standing on the right."
→ "Now move the camera from bottom to top."

The model is said to be capable of simulating physics, gravity and motion, combining Gemini's reasoning with DeepMind's media models (Veo, Genie, Nano Banana).

Personal digital avatars

Omni allows users to create digital avatars bearing their own appearance and voice, which can automatically appear in subsequent videos. To prevent deepfake abuse, the onboarding process requires users to film themselves reading aloud a random sequence of numbers generated by the system — confirming their explicit consent.

⚠️ Current limitations: Clips are limited to 10 seconds (a deployment decision, not a model limitation). The ability to edit voice/audio from uploaded files is currently locked — Google needs more time for safety checks before expanding.

Who can use Omni Flash?

---

Gemini 3.5 — A new generation of language models

Gemini 3.5 Flash — Launched at the event

This model combines frontier intelligence with the ability to perform AI Agent tasks (agentic). Gemini 3.5 Flash outperforms Gemini 3.1 Pro on most programming and multimodal benchmarks, while being 4 times faster than other frontier models in token processing speed. The model is being deployed immediately for the Gemini app, Google Search, Antigravity 2.0 and the Gemini API.

Gemini 3.5 Pro — Launching next month

Currently in internal testing. Built on advanced safety training methods, it minimises harmful content and reduces the refusal of safe questions. Expected to launch publicly in June 2026.

---

Gemini Spark — A 24/7 personal AI Agent

Gemini Spark is positioned by Google as a "proactive AI companion" — not just answering questions but also carrying out tasks on your behalf, running in the background on the cloud even when you are not at your computer.

Spark can:

• Monitor credit card statements, detecting hidden subscriptions or anomalies
• Summarise emails from multiple sources, creating summary documents in Google Docs
• Track notifications from schools, appointments and events
• Connect with third-party apps such as Canva, OpenTable and Instacart

Spark always asks for confirmation before performing important actions such as sending emails, scheduling appointments or completing transactions.

Currently available: Beta for AI Ultra subscribers in the US, rolling out in the coming week. Expanded MCP integration with third-party apps coming in summer 2026.

---

Antigravity 2.0 — A weapon for developers

Google has upgraded its Antigravity development platform to version 2.0, turning it into an AI Agent orchestration hub for developers — a direct competitor to Claude Code, GitHub Copilot and OpenAI Codex.

• Gemini 3.5 Flash runs on Antigravity at speeds 12 times faster than standard benchmarks
• Supports building and managing AI Agent groups to perform multiple tasks in parallel
• Free globally from 19 May 2026
• Integrated into Google AI Studio and Android Studio

---

Ask YouTube & Docs Live — AI enters core products

Ask YouTube

Instead of typing keywords, users can ask complex questions and continue to probe deeper in a chain. The system scans the entire YouTube library — both Shorts and long-form videos — returning the most relevant video list with structured responses.

Currently available for YouTube Premium subscribers in the US via youtube.com/new.

Docs Live

Create and edit documents using natural voice commands, without needing to type AI prompts or make manual edits. Launching on Android and iOS this summer for the AI Pro and Ultra plans.

---

SynthID & Content Credentials — Industry authentication standards

Google is expanding its SynthID digital watermarking system into a cross-industry standard. OpenAI, Kakao and ElevenLabs have already adopted the same C2PA standard. Users can:

• Right-click any image in Chrome and ask Gemini whether it was AI-generated
• View content provenance: captured by a real camera, created by AI, or edited by AI

---

New pricing structure

---

Analysis

Google I/O 2026 is not an event of isolated features — it is a reshaping of the ecosystem. From Omni opening up a path to video creation through conversation, Spark automating personal tasks, to Antigravity 2.0 targeting the AI programming tools market, Google is deploying across all fronts simultaneously.

However, many key features remain geographically restricted (US), tied to expensive plans (Ultra), or are still in the experimental phase. The real question is not what Google can do — but when and for whom it will do it first.

---

Sources: Google Blog, 9to5Google, TechCrunch, The Next Web, Engadget — 19–21 May 2026