Steal these 15 claude code habits (to go from $400 per week to $15...)
1. use haiku for 80% of your work: set haiku as default...it costs 5x less than sonnet and handles bug fixes, file reads, and simple edits just as fast. save $0.80 per session.
2. search first, read second: type "search for x in file" instead of "read file"—searching a 50mb log costs $0.05 vs $5 to read it all. that's 100x cheaper.
3. read files in chunks: say "read lines 1-100" instead of "read entire file"—reading 100 lines costs $0.10 vs $10 for the whole 10mb file.
4. run tasks in parallel: say "run these in parallel" when giving multiple independent tasks—3 files read at once = same cost as 1, but 3x faster.
5. use the explore agent for unfamiliar code: say "explore this codebase for x" instead of random searches...trial-and-error costs $20-30, explore agent finds it first try for $5.
6. plan major changes first (most skip this): say "create a plan for refactoring x" before touching code...costs $0.50 to plan, saves $50 in wasted rework when changing 5+ files.
7. turn on budget alerts: I created a setup script to get automatic warnings at 70% and 90% of your monthly budget. never overspend again.
8. limit search results: ask for "find first 50 matches" instead of "find all matches"...10,000 results cost $5, top 50 cost $0.50, and you only need the first few anyway.
9. be specific in requests: say "fix the login bug in auth.ts line 45" instead of "help me fix this"...vague requests = multiple back-and-forth rounds = 3x cost.
10. use path shortcuts: setup script creates shortcuts automatically so you type utils instead of /src/lib/utilities/helpers/index.ts every single time. faster = cheaper.
11. create task checklists at the start: say "create a todo list for x" at the beginning of multi-step work...prevents forgotten steps and re-reading files, finish 40% faster.
12. read smart: load only what you need: say "read lines 100-200 from database.ts" instead of "read database.ts"...reading 50 lines you need costs pennies vs dollars for all 5,000 lines.
13. don't ask the same question twice: say "use the errors we found earlier" instead of re-running the same search. session memory is free, re-running costs money.
14. let the system filter first: say "show only x from y data" instead of "show all data" get 5 rows with errors instead of loading all 10,000 rows. 95% cheaper.
15. make these habits automatic: I use a setup script to build these into my workflow automatically. install once, save forever.
(real data in the screenshot!)

10,47 mil
153
O conteúdo desta página é fornecido por terceiros. A menos que especificado de outra forma, a OKX não é a autora dos artigos mencionados e não reivindica direitos autorais sobre os materiais apresentados. O conteúdo tem um propósito meramente informativo e não representa as opiniões da OKX. Ele não deve ser interpretado como um endosso ou aconselhamento de investimento de qualquer tipo, nem como uma recomendação para compra ou venda de ativos digitais. Quando a IA generativa é utilizada para criar resumos ou outras informações, o conteúdo gerado pode apresentar imprecisões ou incoerências. Leia o artigo vinculado para mais detalhes e informações. A OKX não se responsabiliza pelo conteúdo hospedado em sites de terceiros. Possuir ativos digitais, como stablecoins e NFTs, envolve um risco elevado e pode apresentar flutuações significativas. Você deve ponderar com cuidado se negociar ou manter ativos digitais é adequado para sua condição financeira.

