@superalesha: I gave DeepSeek an Astra subagent. Then I gave Astra a DeepSeek subagent. Then I made both work alone, because apparent…
Summary
The author experimented by having DeepSeek and Astra AI models work as subagents to build Mario Kart games, comparing the outputs and finding DeepSeek's version more soulful.
View Cached Full Text
Cached at: 09/16/26, 08:06 PM
I gave DeepSeek an Astra subagent. Then I gave Astra a DeepSeek subagent.
Then I made both work alone, because apparently I needed 4 versions of Mario Kart.
Same game prompt, each build from scratch. Astra on medium reasoning, DeepSeek on max throughout. I just wanted to see how different the games would look depending on who was in charge. I still don’t have a good explanation for why I did this.
You can play all the results yourself. Links will be in the first reply.
To be honest, I liked the version best where DeepSeek was the architect and Astra was the implementer. I don’t know why. There’s just more life in what DeepSeek does. Astra’s work looks pretty polished, but it feels kind of soulless. And I love when games have soul and feel alive.
Similar Articles
@aakashgupta: DeepSeek gave an AI agent a team of helpers and 20 jobs to get done, and in five tries it handed off exactly zero of th…
DeepSeek showed that an AI agent given 20 tasks in a 300-message group chat delegated none of them until a two-layer Router/Worker design took the keyboard away from the top agent, cutting slowest-task time from ~30 to under 16 minutes while spending 23.2M cached tokens per run.
@yacineMTB: I'm using deepseek flash 4.1 instead of astra now by the way
A user named yacineMTB tweets that they are using DeepSeek Flash 4.1 instead of Astra, indicating a switch in AI models for personal use.
@VukRosic99: A DeepSeek researcher just open-sourced his AutoResearch personal project. For the first time, the AutoResearch Agent a…
A DeepSeek researcher open-sourced AutoResearch, an autonomous framework that can plan, execute, and debug RL experiments on the DeepSeek 285B model without human intervention, accompanied by a self-play survey paper.
@Hesamation: here's a visual explainer of DeepSeek's article about Super-Agent:
A visual explainer summarizing DeepSeek's article about the concept of a Super-Agent, shared as a link to the explainer graphic.
deepseek-ai/awesome-deepseek-agent
A curated GitHub list of guides for integrating DeepSeek models into popular AI agent and coding-assistant tools like Claude Code, Cline, GitHub Copilot, and more, with installation and configuration walkthroughs.