Open World News

This week's developer updates spotlight two essential pillars of modern AI engineering: hands-on agent building and rigorous experiment tracking. In a new DevDay 2026 session from OpenAI, "Building with GPT-Live and Image Generation," viewers can watch GPT-Live brainstorm ideas, play games, and generate images while coordinating tasks in the background. The video also explores how to add voice capabilities to agents and design effectively for real-world challenges such as delegation, latency, interruptions, and shifting user requests—practical guidance for anyone building responsive, voice-enabled AI systems.

On the evaluation side, MLflow Ambassador Shrinath Suresh continues the MLflow A–Z video series with "E for Experiment Tracking." The episode demonstrates how to track generative AI experiments in MLflow by logging prompts, model parameters, evaluation scores, latency, and token usage, making every run directly comparable. This structured approach helps teams identify which prompt, model, or configuration performs best and iterate with greater confidence.

Together, these posts from OpenAI and MLflow reflect a broader shift toward production-ready AI development—pairing creative, multimodal agent experiences with the measurement discipline needed to improve them. Whether the focus is designing for interruptions in live voice interactions or comparing token costs across model configurations, the message is consistent: successful