12개월 전 알았더라면 좋았을 Claude 활용 팁 11가지
핵심 요약
18개월간 매일 Claude를 사용하며 생산성을 극대화하기 위해 깨달은 실전 활용 팁 11가지.
- 프로젝트 기능 활용 — 코드베이스와 스타일 가이드를 프로젝트 지식으로 등록해 반복적인 컨텍스트 입력을 방지함.
- 커스텀 스타일 설정 — '회의적인 시니어 엔지니어' 같은 페르소나를 설정해 코드 품질을 획기적으로 개선함.
- 모델별 용도 구분 — Sonnet으로 일반 작업을, Opus로 복잡한 설계를, Haiku로 대량의 단순 반복 작업을 처리함.
- Claude Code 활용 — CLAUDE.md와 서브에이전트를 통해 반복적인 설명 없이 병렬 작업을 효율적으로 수행함.
Most "X tips" posts on this sub are surface level. here's the stuff that actually changed how I use claude after 18 months of daily use including 6 months in claude code.
-
The Projects feature is doing more than you think. drop your codebase context, your style guide, your past PRs as project knowledge once. stop pasting the same context every chat. I wasted probably 100 hours before figuring this out.
-
Custom Styles aren't a gimmick. I have one called "skeptical senior eng" that pushes back on my code instead of agreeing with everything. took 3 minutes to set up. single biggest output quality jump I've gotten.
-
Memory is on by default now and it reads your past chats. if your responses suddenly feel weirdly personalized that's why. you can turn it off in settings. (freaked me out for like a week before I trusted it)
-
Search past chats is hidden gold. I forget which chat had the working code. I just ask "what was the final auth setup we landed on last Tuesday" and it pulls it. saves me from scrolling.
-
Sonnet 4.6 is faster than Opus 4.7 and 80% as good for most things. I default to Sonnet now and only switch to Opus for the gnarly architectural stuff. my limit complaints stopped.
-
Haiku 4.5 is genuinely useful for batch work. need to clean 200 support tickets, draft 50 email replies, summarize 30 PDFs. Haiku. don't waste Opus tokens on Haiku tasks.
-
The mobile voice mode is underrated for thinking out loud. I walk for 20 min, talk through a problem, then ask claude to summarize what I'm trying to figure out. solved more decisions on walks than in offsites.
-
In claude code your CLAUDE.md is doing more work than the prompts. write 80 lines of project context once. stop re-explaining your stack every session.
-
Skills > custom instructions for repetitive workflows. I have a skill that pulls the right docs based on what file I'm in. setup took an afternoon, pays off every day.
-
Subagents in claude code unlock parallel work that mostly happens in your head. "spin off a subagent to run the test suite while I keep coding" is the move. most people don't use them at all.
