A practice is something you want to keep doing on purpose, a repeated behavior with a name, not something you did once.
Think about closing a small office at the end of the day. Someone locks the door, someone kills the lights, someone checks the thermostat, someone makes sure the deposits aren’t sitting on the counter. The first time it’s a bunch of little actions. Once the team decides it matters, it becomes the closing practice. A practice differs from a rule: a rule says what must or must not happen (don’t send an email without approval, don’t delete client files), while a practice is softer but still important (run extract after a major agent session, review the output before you send). Rules protect the floor. Practices raise the standard.
How it shows up
Practices often come from noticing the same moment over and over. If something burned once, you fix that instance. If it keeps burning, you change how you cook, and that might become a checklist, a workflow, or eventually a skill. This matters with AI because agents follow whatever pattern you give them. A sloppy pattern just gets faster, while a thoughtful one compounds. “Ask us any questions you need before starting” makes the agent interview you instead of guessing. “Do the work, then have another model review it” adds a judgment gate. That connects to GTD, which isn’t a folder structure but a set of practices around capture, review, and next actions, done often enough that the system earns trust.
Why you care
A practice is how a team learns without pretending everything needs to be formal policy. You name it, teach it, watch whether it works, and then decide whether it deserves to become a rule, a workflow, or a skill. Most good work isn’t one big decision. It’s a small behavior repeated until the team can trust it.