7 tips that turn ChatGPT 6 Astra into AGI

Summary

1. 1.ChatGPT Desktop App and Computer Use. The presenter demonstrates that the desktop application is necessary to access Astra's computer use features, which allow the model to interact directly with web browsers and local software. In an on-screen demonstration, the model navigates Amazon to search for ingredients, select grocery items, and add them to a shopping cart autonomously. Additionally, the desktop interface introduces real-time interactive questions during task execution, letting the model pause and prompt the user for design choices before applying changes.
2. 2.Low Reasoning for Execution and High for Planning. Benchmarks are shown demonstrating that Astra on Low reasoning delivers high accuracy at a lower API cost compared to older models running on high reasoning. The presenter advises using Low reasoning for everyday coding and execution tasks. High, Max, and Ultra reasoning levels should be reserved for initial project architecture, strategic roadmapping, and finding potential blind spots in system design.
3. 3.Eliminating Agent Files and Skill Modules. The presenter argues against using specialized agent rules files and heavy custom skill sets. Because Astra possesses strong core reasoning capabilities, custom skill scripts and rule files often add unnecessary token overhead and bloat the context window. Letting the model determine tool usage dynamically yields faster and cheaper execution without compromising output quality.
4. 4.3D Modeling with Blender and Unity. The video highlights Astra's ability to create 3D assets and game systems. The demonstrated workflow begins with generating concept art via image prompts, converting those designs into 3D models within Blender, and assembling functional gameplay elements inside Unity. The presenter shows a playable shooter demo where weapon models, character logic, and level environments were generated with Astra.
5. 5.Connecting to an External Second Brain. To maintain project memory across long workflows without exceeding token limits, the presenter integrates Astra with a Notion Kanban board. The model drafts project tasks under High reasoning, populates the Kanban columns with structured issue tickets, and subsequently iterates through the tickets on Low reasoning to build and test features sequentially.
6. 6.Remote Mode for Continuous Testing. Astra conducts extensive automated testing cycles that can monopolize local system controls for long periods. To prevent disruptions during daily work, the presenter recommends running the desktop application on an always-on host machine, such as a Mac Studio, and managing tasks remotely from secondary devices like a laptop or phone.
7. 7.Asking Open-Ended Questions Over Direct Prompting. Rather than dictating every micro-step, the presenter recommends asking Astra what steps to take next, where design blind spots might exist, and what high-leverage actions should be prioritized. Allowing the model to formulate suggestions before user approval produces more creative and comprehensive system architecture.