Ian Ballantyne, Developer Relations Engineer at Google DeepMind, demonstrates Gemma 4's multi-agent capabilities by spinning up 10 independent subagents on a single local machine. A high-level prompt goes to an orchestrator, which delegates the work, and within seconds the agents return a complete SVG art gallery, coded locally, simultaneously, and autonomously.
What's covered: Running the Gemma 4 26B model locally, batch processing across 10 concurrent agents, orchestrator-and-subagent task delegation, inference speeds above 170 tokens per second, and what parallel local agents unlock for enterprise workflows and private on-prem deployments.
What would you build with parallel local agents? Drop it in the comments.
Resources:
Github → https://goo.gle/4gzUwd9
Gemma Docs → https://goo.gle/3StWr9e
Gemma Cookbook → https://goo.gle/4vsfKhF
Subscribe to Google for Developers → https://goo.gle/developers
Speaker: Ian Ballantyne
Products Mentioned: Google AI, Gemini
