In Episode #17 of the Agent Factory podcast, we step away from the purely theoretical and get our hands dirty with the latest developer tools from Google. Together with Vlad, we take a deep dive into Antigravity and Nano Banana Pro, demonstrating how to build AI agents that bridge the gap between code generation and high-fidelity media.
This post guides you through the key ideas from our conversation. Use it to quickly recap topics or dive deeper into specific segments with links and timestamps.
Antigravity – a new agentic mission control
Timestamp: 02:02
Antigravity is Google’s new agent-first development application. It is designed as a multi-window IDE that uses Gemini 3 under the hood to manage complex, asynchronous coding tasks. Unlike a standard text editor, Antigravity features a dedicated Agent Manager view where developers can interact with agents through planning modes, artifact reviews, and built-in browsers for live UI testing.
Nano Banana Pro and why it’s so special
Timestamp: 03:47
The model behind our slide generator is Nano Banana Pro (the technical name is Gemini 3 Pro Image model). What sets Nano Banana Pro apart is its ability to “think” before it creates. It uses Google Search grounding to retrieve real-time data, like current weather or live stock charts, and integrates that information into the generated image.
The Factory Floor
The Factory Floor is our segment for getting hands-on. Here, we moved from high-level concepts to practical code with live demos.
Agent Architecture
Timestamp: 06:41
We dove into the architecture for building the slide generator agent using a combination of the Agent Development Kit (ADK), Antigravity, and Nano Banana Pro. Moreover we explained how these components interconnect to support the flow from a user prompt to a high-fidelity presentation by leveraging a Model Context Protocol (MCP) server.
Agent Starter Pack
Timestamp: 10:04
We started by introducing Agent Starter Pack and using it to spin up a new ADK project.
Vibe Coding a Slide Generator Agent
Timestamp: 14:56
Using Antigravity’s “Always Review” mode, we submitted a prompt to build a completely new slides agent which will use a MCP server to create detailed images.
We highlighted how Antigravity handles tasks and implementation plan reviews, allowing the developer to modify the agent’s proposed steps before it executes them. We also dove into automated testing with Antigravity’s web browser plugin functionality.
Results
Timestamp: 32:25
The slides generator agent successfully called the MCP tools to generate a series of images stored in a Cloud Storage bucket, presenting the final links directly in the UI. Antigravity reviewed the results and to our surprise immediately suggested additional improvements.
Conclusion
We are moving into an era where building an agent isn’t just about the LLM; it’s about the orchestration of tools, planning, and high-fidelity output. Antigravity is our agent manager and Nano Banana Pro acts as an asset creator.
We can’t wait to see what our viewers create next with these powerful tools!
Resources and links
-
Antigravity → https://goo.gle/3XMvGwf
-
Nano Banana Pro → https://goo.gle/3KMmKUG
-
Agent starter pack → https://goo.gle/4oLfypT
-
Agent Development Kit (Python) → https://goo.gle/4p0Pszm
-
GitHub with demo code → https://goo.gle/48N83IN
