Tips from seven years of running GTD tools

Hard-won lessons from using, building, and rebuilding GTD software. Take what works.

System hygiene

Capture is cheap, clarify is rare. Drop everything in Inbox the moment it appears. Process it at fixed times (morning + end of day). Don't try to clarify while capturing — you'll stop capturing.
Two-minute rule is real, but strict. Under two minutes, do it now. Longer, capture it. "I'll just do this real quick" is how 90 minutes of unclarified work sneaks into your day.
Your Inbox should hit zero daily, or at least weekly. It's a processing queue, not a todo list. The longer items sit, the worse your trust in the system gets.
Weekly Review is non-negotiable. Every GTD user who quits does so because they stopped doing this one thing. 20 minutes, once a week. Set a calendar reminder if you have to.

Start with a verb. "Write thank-you note to Susan" beats "Thank-you note." The verb forces you to think about what the next physical action actually is.
One task = one action. "Plan birthday party" is a project. The first task in that project is "Make guest list," not "Plan the party."
If you're avoiding a task, it's usually not really the next action. Break it down smaller. Or accept that you don't actually want to do it and move it to Someday.

Every active project needs a Next Action. If it doesn't, it's stalled. The Weekly Review surfaces this. Don't let more than one or two stay stalled through a review cycle.
Name the successful outcome, not the process. "Book is printed" beats "Work on the book." The outcome tells future-you when you're done.
Mark projects sequential when they actually are. Fighting your tool because you think everything should be parallel just makes the Next Actions list noisy. If step 3 genuinely blocks on step 1, say so.

Don't over-tag. A small, used set beats a big elaborate taxonomy nobody maintains. 5-10 tags total is plenty for most people.
Use contexts (@home, @computer, @errands) for filtering, not tags. Contexts answer "what can I do right now given where I am?" Tags answer "what is this about?" They're different dimensions.

If your Someday list has 100+ items, you're using it as a graveyard. Trim it during Weekly Reviews. The list should be "real possibilities I might do" — not "things I'm afraid to say no to."
A stalled project is information. Don't feel guilty about it. Decide: is this really a priority? If not, drop it. If yes, what's actually blocking?
Capture your avoidance. If you keep postponing the same task every week, write down why. Often the real blocker is a missing prerequisite or a wrong framing, not laziness.

AI suggestions are directional, not prescriptive. Let Claude flag things, but trust your judgment. Nothing gets done without your explicit click.
Don't outsource the review itself. The value of Weekly Review is you thinking about your system. AI accelerates the scan, it doesn't replace the decisions.

That's it. Welcome to MLW. May your mind be like water.