Alex Finn opens with the opinion that Astra represents AGI, then focuses on practical workflow examples rather than a formal demonstration of general intelligence. He describes desktop computer useComputer use is an AI capability that interprets a graphical interface and operates software through actions such as clicking, typing, and selecting. for a multi-step grocery-cart task and an interactive development session that asks questions while work proceeds. His first tip is to consider delegation for repetitive computer tasks.
Alex Finn recommends higher reasoning effortReasoning effort is the amount of internal computational work an AI model applies before producing an answer or action. for planning and identifying blind spots, followed by lower effort for execution. He also argues for reducing generic instruction files and skills to avoid unnecessary context, while allowing exceptions for specialized personal workflows. This is his workflow preference and account of model behavior, rather than a verified rule that project instructions or safeguards are unnecessary.
Alex Finn describes a game-asset workflow that begins with generated concept images, builds corresponding 3D assets in Blender and incorporates them into Unity. He then explains how a persistent task board can hold detailed plansTask planning breaks an objective into ordered, bounded steps with dependencies, responsibilities and evidence for completion. from a higher-effort session so that a lower-effort session can work through individual tickets. His claimed cost and speed benefits are experiential rather than measured comparisons in the video.
Alex Finn suggests running development and repeated interface testing on a dedicated computer, with prompts sent remotely from other devices, to reduce interruption on the device being used for other work. His final tip is to ask about blind spots and possible next steps, review the proposed ideas and approve a direction. The recurring theme is separating planning, execution and supervision while keeping useful context available between sessionsAI agent memory is stored information that an agent can retrieve and use across steps, sessions, or changing contexts..
Watch on YouTube
