What Is It With AI Models and In-Browser Racing Games? | ✉️ #99

Hey! 👋

I'll be honest with you: the most impressive thing with GPT-6 Astra so far is how little imagination all of us who use it have.

We got something that can be marketed as AGI, and what do the most demos online look like? Like one more Mario Kart in the browser. I kid you not, I saw at least 10 different "holy shit can't believe Astra did it in one prompt" posts, where the author either re-created another lame version of Mario Kart with WebGL, or created a 3D model that can explode into hundreds of components and re-assemble itself back from this components with a click. Really?

I am sure a lot of people do a lot of actually cool things with Astra. We threw couple of our internal projects at Astra and were quite happy with the result - see a blog post on this quite soon on our website. But it's hard to say with these projects if the result is something unachievable before with Fable or 5.6 Sol, or it's just.. more of the same but maybe a bit better.

What impresses me is when OpenAI says their model solved some unsolvable math problem, or invented a new medicine. But a never ending stream of in-browser games, blender 3D models and re-creation of New York in Unreal Engine (the one city that was already re-created in the most beautiful way in games like Spider Man) just doesn't convince me we are in the AGI age.

We need to share more exciting use cases, more things where it's truly "wow!" and not just "yeah this render of a mansion looks better than what Fable did". Or we might be just hitting the limits of our humanity average imagination, of how far our ideas can go - so we are doomed to just create even more calorie trackers, gym apps, to do lists and habbit trackers, but now faster and with more powerful models.

What We've Shared

• ​The Agentic Engineering Myth - 1 Year of Coding with AI: A practical guide to agentic engineering in 2026, exploring how AI coding agents are changing software development workflows. Learn about tools, context management, architecture decisions, AI skills and plugins, parallel agents, automation, code review, deployment, and the practices that help engineers use AI productively without losing control over their systems.

The 100th mkdev dispatch will arrive on Friday, September 25th. See you next time!