Riley Brown presents seven observations from using GPT-6 Astra in Codex, beginning with a workflow that turns an estate brochure and floor plan into a Blender scene3D scene generation uses AI or procedural tools to create a digital three-dimensional environment from instructions, references, or structured inputs. and then a playable map in a first-person shooter prototype. Riley Brown reports that the model created the initial scene in about 20 minutes and integrated it into the game after roughly an hour, including a flyable helicopter. The example illustrates a sequence across document interpretation, 3D modelling and software development rather than an independent measure of general intelligence.
Riley Brown connects these experiments to a possible expansion of small, customized games and interactive learning tools. Examples discussed include exploded vehicle diagrams and 3D engine illustrations, although Riley Brown explicitly says the engine example has not been checked for technical accuracy. The wider claims about accessible game creation and new hosting businesses are predictions based on these early demonstrations.
Riley Brown also demonstrates a personal desktop utility created to reduce the friction of finding and preparing presentation images. A keyboard shortcut opens image search, background removal, multiple queries and a saved-image collection, with an AI feature splitting a list into separate searches. Riley Brown says the initial application took 16 minutes and one prompt, using the example to explain how narrowly tailored software can address everyday workflow problems.
Riley Brown argues that stronger computer controlComputer use is an AI capability that interprets a graphical interface and operates software through actions such as clicking, typing, and selecting. matters because an agent can inspect and test what it createsAgent self-testing is an AI agent's use of tests or direct inspection to check whether its own work behaves as intended.. In the estate project, Riley Brown describes Codex taking screenshots, reviewing the scene, playing the game, testing helicopter movement and adjusting an unsuitable building texture. These reported iterations provide a concrete account of the build, inspect and revise loopIterative refinement improves an AI result through repeated cycles of building, inspection, feedback, and revision., while the video offers anecdotal demonstrations rather than a systematic reliability evaluationAgent evaluation measures how well an AI agent performs intended tasks across defined, repeatable conditions..
Riley Brown attributes part of the experience to the tools and local application access available through Codex, including browser interaction and access to working files. Riley Brown then outlines a planned always-on computer setup that would keep agent work accessible remotely while travelling. The proposal is still a planned setup, and the claim that these capabilities feel like AGI remains Riley Brown's interpretation of the demonstrations.
Watch on YouTube
