Greg Isenberg defines local AI as running a model on hardware the user controls, from a phone or laptop to an office workstation. He recommends asking whether a smaller model is good enough for a specific job and whether local execution improves privacy, latency, offline access or recurring cost, rather than comparing it only with the strongest cloud model.
Greg Isenberg organizes the ecosystem into four layers: the model, a model repository, runtime software and the workflow built around them. His practical path is to try an open model in a desktop tool, expose it through a local API when needed, test it on a repeated folder-based task and compare its output with a stronger cloud model using a small evaluation. Many useful products, he argues, will use a hybrid design with local processing for sensitive data, cloud escalation for difficult reasoning and human approval for important outputs.
Greg Isenberg proposes three businesses where local execution is central rather than decorative: a quality reviewer for home-health documentation, an offline field-report assistant for restoration contractors and a pre-send reviewer for professional-service firms. Each idea begins with one narrow workflow and a service-led discovery process, then turns repeated mistakes and review criteria into a checklist-driven product.
Watch on YouTube
