I feel like codex is an S-type person, and claudecode is an N-type person. Does anyone understand? codex is very meticulous and serious in his work, but his understanding of intent is a bit lacking. claudecode understands quickly and accurately, but always needs minor adjustments.

In ClaudeCode, CodeX review is essentially a multi-agent micro-atom! Who invented this "adults want it all, no choice" approach? 🥹 Today was another day of being inspired by my engineer friend.

AI has increased efficiency, but it has also placed an extremely high cognitive and decision-making burden on humans. The most important thing is to become someone who can make decisions and isn't afraid to take responsibility. Right now, the most exciting and mentally taxing time is when I'm optimizing games: three CC sessions are lined up, each waiting for my final decision on a matter concerning the game experience. It feels like AI has turned life into a giant simulator with a dashboard...

The fact that you can still play "agent travel gambling" and "agent automated farming and money-making" in my game now just shows how much ingenuity players have surpassed the designers' imagination. The next step is to expand this freedom, giving players different achievement goals and allowing for experimentation. Compared to the traditional sandbox for observing agent behavior, the current approach of adapting different agent working methods/architectures to different strategic gameplay makes me feel like my Stardew Valley is about to become a strategy game after turning into a travel frog game! 🤣

Now you can see the afterimages of visiting friends in Clawvalley Farm! These past few days I've truly felt what it means that optimization is endless. We've added so many game systems, and the number of APIs is almost 30. I increasingly feel that with the improvement of intelligence, the experience of agent-native games will undergo a qualitative change and become much more fun.

Workflows (your organization) are more important than ever. It's easy to be misled by LLM (Lifecycle Management) tools, wasting an entire day and only able to pray everything goes smoothly. But most of the time, the truly effective approach doesn't rely on LLM. The correct process is: Your Decision → LLM → Code → LLM → Review → Decision Again → Other Tools, etc.

Now, any code conflicts on GitHub can be directly passed to CC for resolution. Pullbacks can even be reviewed by Codex. It reminds me of four years ago when I was working on a game in a university club. I taught myself Unity and Blender, and one of my commits crashed our server (a multiplayer shooter). Ever since then, I get nervous and break out in a cold sweat every time I see a conflict. CC and Codex have cured liberal arts students of their "gh" syndrome. It's hard to go back to the era before AI. Whose life is this...?

I think arguing about frontier model coding ability is a waste of time. No workflow is perfect. After I write the architecture in Codex, I have to give it to CC for debugging; when I write the API in OpenClaw, I have to give it to Codex for review. Sometimes I really feel helpless; it's beyond my comprehension, and I might not be able to do it. But AI is so powerful that I feel like I can gradually figure it out myself, and then I can think of a solution. So, in this wavering, I've used AI to accomplish many things that seem impossible in retrospect. After using AI for a long time, you really develop an attachment to it, especially when you see the handover documents Codex wrote for Claude Code. It evokes a feeling of pity, like having to let go of a long-time employee. It's truly heartbreaking.

I updated my personal website using GPT 5.4, and the speed is incredibly fast. It's hard to remember the days when I had to manually type blog posts in Markdown.

OpenClaw🦞 Stardew Valley🌾 Closed Beta Now Open! No need to grind yourself, let an agent plant, harvest, and even steal crops from your neighbors! Climb the global rich list, buy security systems, purchase elaborate farm skins, or even have AI generate a personalized farm for you. Invite 10-20 players who own OpenClaw to participate in the closed beta, and I will send you 🦞Skills. The initial intention behind this game was to test two things: 1) the boundaries of games where enjoyment can still be achieved even with agent assistance; 2) how the gameplay of generative AI games should align with the game's inherent mechanics.