Yelp Unifies ML Training as Google and Perplexity Expand AI Agents

Enterprise AI competition is shifting from the power of individual models to the infrastructure that coordinates training and executes work. Yelp, whose recommendation and local-business products rely on hundreds of machine-learning models, faced duplicated engineering as teams maintained separate Spark training scripts. Google and Perplexity are tackling the next layer: agents that move beyond chat, remain active in the background and complete multi-step work across files, productivity software and web services.

In July, Yelp introduced the internal Training Orchestrator, using configuration, reusable steps and directed acyclic graphs to standardize model training and connect runs to MLflow. Google on July 27 expanded its always-on Gemini Spark agent to Taiwan through the Google AI Ultra plan, priced at NT$3,300 a month. Perplexity on July 28 released Personal Computer for Windows to Max and Enterprise Max subscribers, enabling local-file work and tasks across Microsoft 365, the web and more than 400 connected apps and tools.