By Google I/O 2026 starts tomorrow. Before the keynote floods your feed with consumer demos and “AI-powered” widget reveals, here is the part worth paying attention to: Jules is Google’s direct answer to Claude Code and Codex, ADK 1.0 is the only major agent framework that ships native Java and Go support, and Veo 3 is now in the Gemini API at pricing that developers can actually consider. There is real signal in this I/O. Here is where it is.
Jules: Google’s Async Coding Agent Finally Gets Serious
Jules is not an IDE plugin or a chat assistant. It is an asynchronous coding agent that spins up a cloud VM, clones your repository, writes a plan using Gemini 3 Pro, runs your tests, and opens a pull request — all while you do something else. That is the key difference from synchronous agents: Jules is designed for parallel execution. Submit ten tasks simultaneously and come back to ten PRs.
The Gemini CLI integration brings Jules into your terminal without switching tools. Install the extension and delegate tasks with /jules [task description] from within Gemini CLI. The VM runtime includes Node.js, Python, Go, Rust, and Java — not limited to TypeScript shops.
# Install Jules extension for Gemini CLI
gemini extensions install https://github.com/gemini-cli-extensions/jules --auto-update
# Submit an async task from the terminal
/jules Fix the race condition in the payment processor and add testsThree separate I/O sessions are dedicated to agentic coding. That is not accidental — Google is making Jules the headline of this event. Whether it earns that depends on execution. Google has a documented history of building excellent developer tools and quietly sunsetting them. That skepticism is fair to hold while watching the demos. Visit jules.google to get on the waitlist.
ADK 1.0 GA: The Multi-Language Moat
Google’s Agent Development Kit hit 1.0 GA across Python, TypeScript, Java, and Go. LangGraph, CrewAI, and the OpenAI Agents SDK are Python-first — often Python-only. If your team runs Java microservices or Go backend services, ADK is currently the only mainstream agent framework that does not require a Python dependency to ship agentic workflows.
ADK Python 2.0 (beta) adds the AgentTeam API for multi-agent coordination. The A2A (Agent2Agent) protocol treats tasks as first-class objects with Server-Sent Events for progress streaming — no more polling on long operations. In production benchmarks, ADK’s Event Compaction cut token usage by 38% and improved latency by 18% by summarizing older context instead of truncating it.
| Framework | Languages | Wins On | Best For |
|---|---|---|---|
| Google ADK | Python, TS, Java, Go | Cost, language breadth | GCP-native teams |
| OpenAI Agents SDK | Python | Developer experience | Python shops, quick start |
| Claude Agent SDK | Python, TS | Reasoning quality | Complex logic chains |
| LangGraph | Python, JS | Ecosystem size | Complex orchestration |
Veo 3 in the Gemini API: Video as a Build Primitive
Veo 3 is now in the Gemini API at pricing worth paying attention to. Veo 3.1 Lite runs at $0.05 per second of generated video — the first AI video API where an independent developer can prototype without burning a monthly budget in an afternoon. Veo 3.1 Standard at $0.40/second adds native synchronized audio and higher fidelity. Both support text-to-video and image-to-video, with native 9:16 output for Shorts and Reels.
The image-to-video path is the more interesting developer primitive: provide a reference image alongside a text prompt to maintain consistent characters and scenes across generated clips. Useful for product content, educational explainers, and app onboarding without manual video production. Access requires the paid Gemini API tier.
Gemini 4 and the 2M Token Context Window
Google has not officially announced “Gemini 4” by name, but session signals and pre-I/O leaks point toward a major model update with a 2 million token context window — double the current Gemini 3.1 Pro ceiling. At 2M tokens, a large codebase fits in context without retrieval augmentation. That matters for Jules specifically: an agent that reads your entire codebase will produce fewer nonsensical diffs than one sampling it.
Hold the excitement calibrated to one question: what does it cost per token at 2M? Gemini 3.1 Pro at 1M is already expensive at scale. The spec is interesting; the pricing is what determines developer utility.
Android 17: The Forced Layout Migration
Android 17 (API 37) removes the orientation and resizability opt-out for large screens. Target API 37 and your app must handle adaptive layouts — there is no escape hatch. On the upside: a new lock-free MessageQueue implementation reduces UI jank, and local network access is blocked by default (use ACCESS_LOCAL_NETWORK for persistent access). Final release is expected around July 2026. Start your layout migration before the summer crunch.
The Real I/O 2026 Story
Google is not just shipping models this year. It is shipping infrastructure: Jules for async cloud VM coding, ADK for multi-language agent orchestration, Veo 3 for video generation in the API, and a unified OS strategy under Android 17. OpenAI has GPT-5.5 and Codex. Anthropic has Claude Code and MCP. Google’s answer is to own the full stack from the VM running Jules to the ADK orchestrating agents to Firebase as the agent-native backend. The developer keynote tomorrow at 1:30 PM PT is where we find out if the execution matches the ambition.
