Exploring the Edges of AI Art with Hermes

Genesis

It began four months ago, when "Lobster" suddenly took off, and it has kept evolving into what is now Hermes.

The truth is, most of the time — even though Hermes has been running on the server this whole time (it used to be "Lobster") — I barely touched it. Most concrete tasks felt like they could go straight to Codex and Claude Code. If you need a scheduled, more granular, freezable workflow, you can just have those two generate the project code and carry the task out directly.

For me, both Openclaw and Hermes started to feel a little superfluous. Part of it is probably budget: once Anthropic rolled out its subscription plans and siphoned Openclaw off, almost no one wanted to wire Openclaw up through Anthropic's API anymore (deep pockets aside).

Is Artistic Thinking a Way of Escaping the Subject?

Recently I happened across a friend's idea of using AI to make generative art. On a whim I tweaked the prompt, hoping that, through continuous refinement, my AI could probe the edges of that space too — and explore more of what's possible in language as well.

Once GPT and Codex can be wired in too (that's the precondition), and once it turns into a kind of generative art, you no longer have to be bound by its subject or its function. It's as if the meaning of its existence suddenly grew much richer.

So, for now, let my agent take these two dimensions — Agent Art and Agent Thoughts — as its first main task.