Michael Shimeles discusses practical GPT-6 Astra use cases with Greg Isenberg, beginning with software performance reviews, security audits and interface design guided by reference images. The conversation also considers computer useComputer use lets an AI agent interact with software through interface controls such as screens, buttons, keyboards and pointers. for navigating websites and gathering information. Reported improvements are personal examples rather than controlled comparisons or evidence that an automated review establishes security.
Greg Isenberg proposes breaking service businesses into explicit stepsWorkflow automation uses software or AI to complete a connected sequence of routine steps with less manual effort., required tools, inputs, outputs and points needing human judgmentHuman-in-the-loop AI keeps a person involved in reviewing, correcting, approving, or guiding an AI system's work. before designing an AI product. Other examples include an operating dashboard built from company documents and analytics, identifying tasks suitable for agents, documenting browser workflows and reviewing applications for both defects and confusing user experiences. These are proposed workflows, not proof that every task has been automated successfully.
Michael Shimeles presents a plan for a Raspberry Pi speaker connected to a personal AI agentA personal AI agent uses a person's authorized goals, preferences, routines, context, and tools to provide continuing individualized assistance.. The walkthrough describes using Blender to illustrate the component layout, assembling a parts list and preparing repository changes through a pull request. The equipment still needed to be assembled and tested at the time of the discussion, so the example demonstrates planning and code preparationTask planning breaks an objective into ordered, bounded steps with dependencies, responsibilities and evidence for completion. rather than a verified working device.
Greg Isenberg connects the hardware example to exploring small physical products and learning unfamiliar technical skills. Michael Shimeles also considers repurposing an unused game console as a Linux learning project, but no completed installation is established. The practical theme is to use iterative investigation, visual explanations and concrete prototypes to explore a project beyond a familiar software niche.
Watch on YouTube
