Uploaded August 2026 | Updated September 2026, 3 weeks ago
Roadmap preview → https://goo.gle/mma-roadmap
Most AI agents are text in, text out. This AI agent hears you, sees the page, remembers context, talks back, and moves a real browser: live, while you're still speaking.
I call that an omni-app: perceive → reason → express, running as one open loop instead of taking turns. In this video, Annie breaks down what a bidirectional live multimodal agent actually is, why it's a much bigger jump than "a better chatbot," and the full 6-episode roadmap for building one yourself with the Gemini Live API and Google's Agent Development Kit (ADK).
🗺️ *SEASON 1 — THE ROADMAP*
* Episode 1 · Voice — text-to-speech vs. a live session, and why interruption changes the whole interface
* Episode 2 · Framework — raw GenAI SDK vs. Google ADK: who runs the loop
* Episode 3 · Tools — function calling inside a live conversation, and the latency problem it creates
* Episode 4 · Browser — giving your agent hands: voice-controlled browser automation
* Episode 5 · Memory — structured facts + vector memory, retrieved mid-conversation
* Episode 6 · Vision — one snapshot vs. a continuous frame stream, without killing the live feel
⏱️ *Chapters:*
0:00 - The demo: talking to an agent that actually acts
0:50 - What is an omni-app?
1:35 - Chatbot vs omni-app: one turn vs one live loop
2:10 - What "bidirectional" actually means
2:50 - The loop: Perceive → Reason → Express
3:25 - Season 1: six episodes
3:55 - Episode 1 · Voice — text-to-speech vs live (and interruption)
4:30 - Episode 2 · Framework — GenAI SDK vs Google ADK
5:00 - Episode 3 · Tools — and the tool-latency problem
5:40 - Episode 4 · Browser — giving the agent hands
6:00 - Episode 5 · Memory — facts, vectors, and live recall
6:20 - Episode 6 · Vision — snapshot vs continuous frames
6:55 - What developers will be able to build
🔗 *Resources:*
* Roadmap preview:→ https://goo.gle/mma-roadmap
* Code for this series → https://goo.gle/mma-code
* Gemini Live API docs → https://goo.gle/4xxUoQD
* Agent Development Kit (ADK) docs → https://goo.gle/4g6JH0n
Connect with Annie online:
LinkedIn → https://goo.gle/annie-linkedin
X → https://goo.gle/annie-x
Watch more The Omni App → https://g.dev/cloud/modern-ai-app
🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech
#AIAgents #GeminiAPI #MultimodalAI
Speaker: Annie Wang
Products Mentioned: GenAI SDK, Google Agent Development Kit, Gemini Live API
Roadmap preview → https://goo.gle/mma-roadmap
Most AI agents are text in, text out. This AI agent hears you, sees the page, remembers context, talks back, and moves a real browser: live, while you're still speaking.
I call that an omni-app: perceive → reason → express, running as one open loop instead of taking turns. In this video, Annie breaks down what a bidirectional live multimodal agent actually is, why it's a much bigger jump than "a better chatbot," and the full 6-episode roadmap for building one yourself with the Gemini Live API and Google's Agent Development Kit (ADK).
🗺️ *SEASON 1 — THE ROADMAP*
* Episode 1 · Voice — text-to-speech vs. a live session, and why interruption changes the whole interface
* Episode 2 · Framework — raw GenAI SDK vs. Google ADK: who runs the loop
* Episode 3 · Tools — function calling inside a live conversation, and the latency problem it creates
* Episode 4 · Browser — giving your agent hands: voice-controlled browser automation
* Episode 5 · Memory — structured facts + vector memory, retrieved mid-conversation
* Episode 6 · Vision — one snapshot vs. a continuous frame stream, without killing the live feel
⏱️ *Chapters:*
0:00 - The demo: talking to an agent that actually acts
0:50 - What is an omni-app?
1:35 - Chatbot vs omni-app: one turn vs one live loop
2:10 - What "bidirectional" actually means
2:50 - The loop: Perceive → Reason → Express
3:25 - Season 1: six episodes
3:55 - Episode 1 · Voice — text-to-speech vs live (and interruption)
4:30 - Episode 2 · Framework — GenAI SDK vs Google ADK
5:00 - Episode 3 · Tools — and the tool-latency problem
5:40 - Episode 4 · Browser — giving the agent hands
6:00 - Episode 5 · Memory — facts, vectors, and live recall
6:20 - Episode 6 · Vision — snapshot vs continuous frames
6:55 - What developers will be able to build
🔗 *Resources:*
* Roadmap preview:→ https://goo.gle/mma-roadmap
* Code for this series → https://goo.gle/mma-code
* Gemini Live API docs → https://goo.gle/4xxUoQD
* Agent Development Kit (ADK) docs → https://goo.gle/4g6JH0n
Connect with Annie online:
LinkedIn → https://goo.gle/annie-linkedin
X → https://goo.gle/annie-x
Watch more The Omni App → https://g.dev/cloud/modern-ai-app
🔔 Subscribe to Google Cloud Tech → https://goo.gle/GoogleCloudTech
#AIAgents #GeminiAPI #MultimodalAI
Speaker: Annie Wang
Products Mentioned: GenAI SDK, Google Agent Development Kit, Gemini Live API
