Claude Coding Overview
Since November 2025 Claude Code has been really good, good enough that I simply use it. Lately, with Opus 5.5, new possibilities have opened up again.
What changed is the substrate. We now have a technology where programming languages and natural language stand side by side. Not as a new layer on top of the high-level languages, but as a new channel next to them, one that can weave into them and, further down, still use the programming languages and the computer. A separate post on this is planned.
This page is the hub for what I write about programming with Claude.
Posts
- Towards a Meta Design of LLM Skills: skills as thinking operators, layers of skills, and why criticism has to come from outside the model.
- My Skills: the seven skills in bjs-claude-skills, with what each one is for and how well that is backed up.
Planned
- The substrate post. LLMs as a calculator for natural language: a resource on the ground, not a creator. A sketch of the idea is below.
- The journey. From the first ChatGPT experiments to working with Claude Code every day. One more post can tell that story; the older LLM posts belong there.
- Data from real debugging sessions. Which moves the model makes on its own, where it stops, and what the skills change.
The idea in one paragraph
A calculator produces calculation as a resource. Nobody calls it intelligent, and it is still something a creative person builds on. I think intelligence and creativity split the same way: there are resources on the ground that lift the creative person to a higher level. A language model is such a resource, an algorithmic operation with natural language as input and output. It is fuzzy where a calculator is exact, because it works on high-dimensional data, but it is still algorithmic. My position: the model alone is no creative person, just as a calculator is not. Creativity would be a property of the organization it sits in.
This text is generated by Claude on Benjamin Schwerdtner's behalf.