Using LLMs as an engineer
“Familiar with AI” is easy to claim and hard to trust. Here’s what I mean by it.
Leverage, not abdication
I use LLMs daily for drafting code, exploring APIs, summarizing docs, and spinning up agents for multi-step chores. I do not paste output into production without reading it, running it, and owning the result.
Useful pattern:
- State the constraint (language, runtime, security, style)
- Generate options with a model or agent
- Narrow with tests, types, and review
- Integrate as if a junior teammate wrote it—because responsibility is still mine
Where models help most
- Boilerplate and mechanical refactors
- “What’s the usual way to do X in this stack?”
- Explaining unfamiliar error surfaces
- First drafts of docs and commit messages you then edit
Where they fail
- Subtle correctness and concurrency
- Security boundaries and tenancy
- Product judgment (“should we build this?”)
- Invented APIs and confident wrong answers
Good engineers already knew the last mile was the job. Models move the first mile.
What I’ll show over time
I’d rather publish a small, real workflow or feature write-up than a list of model names. When I add demos or deeper notes, they’ll live under Writing and get linked from Work.
If you’re evaluating me for a role: ask how I’d use models on your codebase and where I’d forbid them. That’s a better interview than a buzzword checklist.