Google's Biggest AI Announcements (I Was There)

May 22, 2026

AI Summary

5 min read

Google's IO event centered on shifting from general AI models toward agentic systems capable of handling long-running, multi-step tasks. The discussion between the hosts and Logan Kilpatrick focused on new model releases, their distribution, and concrete ways developers and founders can integrate them into products or content workflows.

Gemini 3.5 Flash and the agent workhorse model

Gemini 3.5 Flash is positioned as the primary model for agentic workloads rather than simple chat. It combines coding performance previously associated with larger models while remaining inexpensive to run at scale. The model is already live across Search for 900 million users, the Gemini app, and the API, creating immediate distribution that earlier releases lacked.

Distillation techniques allow Flash to inherit reasoning and tool-use capabilities from larger models without proportional increases in cost or latency. Builders are encouraged to test it directly against Sonnet-class models and report gaps, since the team treats feedback as direct input for iteration. The practical implication is that agent loops involving repeated tool calls or extended coding sessions can now run continuously at lower marginal cost.

Gemini Omni for cross-modal content and editing

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

Listen to Audio Summary Open in App

What you'll learn

1 (00:10) **Google IO Focus on Agents** - Episode goal is covering new releases with emphasis on agent-native product building
2 (00:53) **Gemini 3.5 Flash Launch** - New workhorse model positioned for long-running agentic tasks and coding
3 (01:54) **Flash Positioning vs Competition** - Anchored at Sonnet-level intelligence, available day one to 900M+ users and developers
4 (02:45) **Gemini Omni Introduction** - New unified multimodal model handling any input to any output
5 (04:17) **Creator Use Cases for Omni** - Enables rapid ad, commercial, and social content generation at scale
6 (06:24) **Product Opportunities Around Omni** - Opens new category for video remixing and editing tools
7 (08:53) **3.5 Pro and Distillation Plans** - Pro variant expected early next month; distillation pushes frontier intelligence into Flash

+ Full timestamped outline available in the app

Show Notes

Live from Google I/O, I sit down with Logan Kilpatrick from the Google DeepMind team to unpack everything Google just announced and what it means for founders, developers, and anyone trying to build with AI right now.

We dig into Gemini 3.5 Flash, the new Gemini Omni world model, the expanded Antigravity ecosystem, managed agents in the Gemini API, and the native Android app builder inside AI Studio. Logan breaks down how distillation is pushing Pro-level intelligence into Flash, where the biggest opportunities are for solo founders, and why the agentic era has finally moved from impressive demos to genuinely useful products.

Thanks to Google for flying me out to Google I/O and making this conversation possible.

Timestamps

00:00 – Intro

00:53 – Gemini 3.5 Flash: The New Workhorse Model

01:49 – How Flash 3.5 Stacks Up Against Sonnet

02:38 – Gemini Omni: A World Model for Any Input and Output

06:18 – Building a Content and Creator Layer on Omni

08:21 – What to look forward to

10:53 – Google Spark and Managed Agents

14:00 – The Agentic Era and Requests for Startups

17:17 – The Antigravity Ecosystem Overhaul

18:51 – AI Studio vs. Antigravity: Vibe Coding vs. Agentic Engineering

21:31 – Native Android Apps Built Inside AI Studio

23:44 – Closing Thoughts

Key Points

Gemini 3.5 Flash ships as a Sonnet-level workhorse model tuned for long-running agentic tasks, coding, and tool use, available on day one to 900M+ Gemini app users.
Gemini Omni is a single model that takes any input and produces any output across video, image, audio, and music, fusing Veo, Nano Banana, Lyria, and TTS into one system.
Managed agents in the Gemini API let builders ship agentic products with a single API call, using skills and markdown instead of writing orchestration code.
The Antigravity suite now spans an IDE, agent manager, CLI, SDK, and API surface, all sharing the same agent harness that powers Gemini Spark.
AI Studio targets vibe coding and now builds native Android apps for free, while Antigravity targets production-quality, million-line-codebase engineering.
The cost of intelligence keeps dropping thanks to distillation, opening up smaller markets that previously needed a 40-