Back to Lab
Lab1 min read
Experiment: a Claude Code workflow
Notes on bringing Claude Code into the development workflow: goals, setup and the questions to answer.
Goal
Find out which parts of the development process an AI coding agent can handle reliably, and which still need tight human control.
Setup
- Test project: this personal website itself (Next.js, TypeScript, MDX).
- Every task has a clear goal, constraints and definition of done.
- Every change must pass type-check, lint and build before it is accepted.
Questions to answer
- Which kinds of tasks does the agent handle well: scaffolding, refactoring, writing content or fixing bugs?
- How does the level of detail in a request affect the quality of the result?
- What are the common failure modes and how can they be caught early?
Status
In progress. Results and lessons will be added to this post.