ChatGPT Thinks With You. Claude Builds With You.
I switched from ChatGPT to Claude. The difference isn't the model. It's the shift from cognitive support to digital crafting, and what that makes possible.
ChatGPT made me think better. Claude makes me build differently. That distinction took a few weeks of actual use to understand, and it's the reason I switched.
With ChatGPT, the dynamic was cognitive support: a sharper search, a thinking partner, a way to process ideas faster. Useful, often excellent. But the work still lived with me. The tool assisted; I produced.
Claude changed the direction of that relationship. The environment composes.
Teaching, not prompting
The mechanism is project files, CLAUDE.md files, and skills. You invest time once in explaining your context, your style, your workflow logic. The system holds it. Every subsequent session starts from that foundation rather than from scratch.
That investment compounds. I have a skill for drafting posts on this blog, one for processing Schmuki meeting transcripts, one for bookkeeping in MoneyBird. Each encodes a workflow I would otherwise rebuild conversation by conversation. The result is consistency at a pace I couldn't sustain manually. One person, producing at team scale.
This is different from prompting. Prompting is asking. Teaching the environment is closer to hiring: you explain how things work here, and the system operates accordingly.
Claude Code and the Git prerequisite
Claude Code added another layer. It's a terminal-first coding environment, agentic in a way the web interface isn't. I used it to build a treemap of Dutch employment and AI exposure, pulling public data, structuring it, iterating on the visualisation through conversation rather than through manual edits. The kind of data project that previously required either a developer or a very patient afternoon.
What I hadn't expected: the unlocking skill wasn't prompt technique. It was Git. Moving a WordPress site to GitHub, learning version control well enough to work with it fluently, that turned out to be the prerequisite for getting real value from Claude Code. The model handles the code. Git handles the ground beneath it.
What digital crafting feels like
There's a term forming for what this is: digital crafting. Not using AI as a tool, and not being replaced by it. Building with it, encoding your practice into the environment, and operating at a scale that was previously out of reach for a single practitioner.
The question I'm sitting with: as more of the workflow gets encoded and more of the output gets consistent, where does the craft actually live? I think it lives in the teaching. In knowing what to encode, how to instruct, where to leave room for judgment. That's not a diminished form of work. It's a different one.
