SYSTEM INSTRUCTIONS FOR GEMINI 3 PRO

Summary from the TLDR AI newsletter

This post provides system instructions for Gemini 3 Pro that improved performance on various agentic benchmarks by up to around 5%. Specific instructions are often required to control how the model reasons, plans, and executes tasks for deep agentic workflows. Complex agents often require a trade-off between computational cost (latency and tokens) and task accuracy. A guide on how to build better agentic workflows is linked in the thread.