Artemis Returns, AI Compute Wars, and Codex Control
Monday, April 20th, 2026
Artemis gets a victory lap as the crew celebrates the mission’s safe splashdown and talks about how a future moon landing would dominate the internet in a way Apollo never could. From there the conversation turns into an extended AI state-of-the-industry check-in, focusing on Anthropic’s reported compute bottlenecks, Claude reliability complaints, and the restricted Mythos model that appears powerful but not yet practical to serve widely. They compare Anthropic’s strategy with OpenAI’s emphasis on efficiency, lower-cost coding performance, and upcoming model releases, while also discussing how AI companies are navigating government and defense relationships. The back half becomes a hands-on look at OpenAI’s Codex computer-use features, with examples ranging from inbox summaries and printed morning briefings to media sorting, podcast post automation, and desktop app control, all framed around the idea that AI works best when you identify which parts of a workflow require human taste and which parts are just repetitive clicking.
Picks:
Andrew Mayne: Astromat YT
Justin Robert Young: Defunctland’s video on the broken promise of Disney intelligent characters
Brian Brushwood: The pilot of Magnum P.I.
Andrew Mayne: Double Reel TV
Support Weird Things on Patreon
Subscribe to the Weird Things podcast on iTunes Podcasts
Podcast RSS feed
Download url: https://weirdthingspodcast.com/uploads/podcastr/8faf3d1f-528a-444e-8218-ba1f24be55a0__260420_-_weird_things_podcast.mp3
[podcast]https://weirdthingspodcast.com/uploads/podcastr/8faf3d1f-528a-444e-8218-ba1f24be55a0__260420_-_weird_things_podcast.mp3
OpenAI’s shutdown of the Sora app kicks off a broader discussion about how AI companies are being shaped less by hype cycles than by raw compute limits, with Disney deal fallout, Anthropic’s work-hour throttling, and rumors of even bigger next-generation models all pointing to infrastructure being the real bottleneck. From there, the conversation shifts into what these tools look like in practice: Andrew talks through using Codex, plugins, and repeatable evals to automate work, build tiny playable games under extreme constraints, and treat coding more like cultivating projects than manually assembling software line by line. The hosts compare notes on how intimidating the current tool landscape can still be for newcomers, why iterative prompting and experimentation matter more than waiting for a perfect “super app,” and how app stores may be poorly equipped for a wave of AI-generated software. They also detour into social media, scams, platform incentives, and the question of whether better guardrails earlier on could have reduced some of the worst outcomes of the last platform era before wrapping with movie, parenting, and gadget recommendations.