Google Gemini 3.5, Omni, and Managed Agents: A Deep Dive
Greg Isenberggo watch the original →
the gist
Logan Kilpatrick from Google DeepMind breaks down the shift toward agentic AI, the release of Gemini 3.5 Flash, and the new managed agent capabilities in the Gemini API.
The Shift to Agentic Workflows
Logan Kilpatrick emphasizes that the current AI landscape has moved beyond simple chat interfaces into the 'agentic era.' The focus is no longer on impressive but fragile demos, but on building reliable, long-running agents that can handle complex, multi-step tasks without constant human intervention. This shift is supported by new infrastructure that allows developers to build agentic products with minimal orchestration code.
Gemini 3.5 Flash and Model Distillation
Gemini 3.5 Flash is positioned as a high-performance 'workhorse' model. Kilpatrick explains that through model distillation—a technique where the intelligence of larger models is compressed into smaller, more efficient ones—Google has achieved Pro-level capabilities in a Flash-sized package. This model is specifically tuned for coding, tool use, and agentic tasks, and is now available across Google’s massive ecosystem, including the Gemini app and API.
Gemini Omni: The Unified World Model
Gemini Omni represents a fusion of Google’s specialized models (Veo for video, Nano Banana for image, Lyria for music, and various TTS systems) into a single, multimodal world model. This allows for seamless input and output across text, audio, video, and image formats. Kilpatrick suggests this will fundamentally change content creation, enabling creators to produce high-quality, complex media without needing large production teams.
Managed Agents and Developer Experience
To lower the barrier to entry for building agentic applications, Google introduced managed agents in the Gemini API. Developers can now define agent 'skills' using simple Markdown rather than writing complex orchestration logic. This system shares the same harness as Gemini Spark, allowing builders to deploy production-ready agents that handle tool calling and task execution with a single API call.
The Antigravity Ecosystem
Google has overhauled the Antigravity suite, transforming it from a simple IDE into a comprehensive ecosystem. It now includes an agent manager, CLI, SDK, and API surface. The goal is to meet developers where they are—whether they prefer 'vibe coding' in AI Studio or building production-grade, large-scale systems via the Antigravity CLI and SDK.