AI coding assistants have changed how I work. Not by replacing my judgement, but by removing friction from the parts of software development that are tedious rather than creative.
What AI is Good At
Boilerplate. Writing CRUD endpoints, form validation, database migrations — tasks where the pattern is clear but the typing is repetitive. I describe what I need; the AI produces a first draft; I review and adjust.
Translation. For this site, I write content in English and ask Claude to produce Malay and Chinese versions. I review the output with native speakers, but the first draft is handled.
Explaining unfamiliar code. When I encounter a library or pattern I haven’t used before, having an AI explain it interactively is faster than reading docs end-to-end.
What AI is Not Good At
Architectural decisions. The AI doesn’t know your constraints, your team, or your users. It will suggest reasonable patterns, but the decision still requires human judgement.
Correctness guarantees. AI-generated code is often subtly wrong in ways that aren’t obvious. Tests, code review, and running the actual software remain essential.
My Workflow
I use Claude Code for longer tasks where context needs to persist across multiple files. For quick questions or snippets, I use the Claude web interface.
The key is treating AI output as a draft, not a final product. Review everything.