Google Research has introduced the Planetary Prediction Engine (PPE), an experimental AI system that autonomously executes the full geospatial modeling workflow—from data discovery and feature engineering to model training and evaluation. The system accepts natural-language queries and produces predictive models without manual intervention, compressing what typically requires weeks of specialized work into minutes. PPE operates through three LLM-orchestrated stages: intelligent geospatial data selection that retrieves relevant covariates from established repositories and conducts live web discovery, multimodal dataset curation that fuses satellite imagery embeddings and demographic foundation models while enforcing automated safeguards against data leakage, and automated model building that searches across multiple architectures with overfitting protection. The modular design passes data between stages via opaque handles rather than serializing into prompts, avoiding context-window limitations. Testing across diverse applications shows significant improvements over manual expert pipelines. On CDC health indicators, PPE achieved a mean R² score of 76.8% versus 60.0% for traditional methods; it similarly outperformed baselines on FEMA risk prediction and vulnerability mapping. For food security analysis in data-scarce regions, PPE doubled baseline accuracy when downscaling from provincial to local government area levels, offering humanitarian organizations rapid, high-fidelity vulnerability assessment during crises.