Statement on AI useage

I increasingly use Claude Code CLI (by Anthropic) as a coding assistant, similar to how a machinist might use CNC tools - Increasingly automatic, carefully monitored, with human interaction required at certain touchpoints, retools, or inspection points; and a lot of work front-loaded in design. There is often less manual, flow state craftsmanship except on the most critical parts.

I find high value in Claude Code CLI for:

• Being a “smart search engine”
• Generating \emph{initial} boilerplate code (e.g., CI/CD pipelines, config files)
• Organizing notes and ideas
• Oneshotting small code snippets, glue modules, pipelines, or even whole CLI devtools

One especially useful workflow is to generate a CLI regression test from a specific input/output example. My time spent on this activity has dropped dramatically since I started using Claude for it, with no impact to quality.

I find it less useful for:

• Writing core algorithms or critical subsystems (I sleep better when I do this myself)
• Any areas I want to deeply understand or have full control over

I adore making real, useful software by hand, so I still tend to turn off all the tools, go internet-free, and focus deeply for side projects, challenge sets, or when required for critical components.

CNC Machining fundamentally changed the trade, requiring increasing “white collar” skills, not less. It massively increased the output a machinist can have, but also increased their responsibilities. This is the best analogy I can find for LLM agents.