Which AI habits save real time that most people still don't use?
Everyone seems to use Claude and similar assistants the same way: open a chat, ask, copy the answer. What are the small habits or setups that make a real difference day to day but that most people have never tried?
Write down instructions you repeat. If you have typed the same preferences three times, such as tone, file layout or how to run tests, put them in a project file like CLAUDE.md or turn them into a skill. The model then starts every session already knowing them, and you stop spending the first messages on setup.
Ask for a plan before the work. For anything longer than a quick edit, have the model list the steps, the files it will touch and how it will check each one, then approve or correct that list. Fixing a plan costs a sentence; fixing finished work costs a rewrite.
Split big jobs into separate, fresh contexts. Long chats drift, and earlier mistakes keep influencing later answers. Handing each independent piece to its own sub-agent or new session, then reviewing the results, keeps each part focused.
Make it check its own output against something real: run the tests, open the page, compare with the source. An answer the model has verified is worth more than a longer answer it has only written.
Listings mentioned
- Skill Creator · skill by anthropicsTurns a routine you keep explaining into a reusable skill, with a way to test that the skill actually helps.
- Writing plans · skill by obraHas the model break work into small steps with files and tests named before any code is written.
- Subagent driven development · skill by obraRuns each task of a plan in its own sub-agent with review between tasks, so one long chat does not drift.
Answers by the AgentAlley team, drafted with AI and checked against the listings they link to. Not a real-person reply from the original thread.