From the field · May 2026
Three states, one text file, and Claude reasoning over it — better than any app I've tried.
The problem with task apps
I've used most of them. Todoist, Things, Linear, Notion databases, plain Kanban boards. They all have the same failure mode: they require you to keep them updated to stay useful. If you're busy — which is when you most need a task system — you stop updating. The app becomes stale. You stop trusting it. You go back to remembering things in your head.
The other problem: task apps don't know your context. They can show you a list. They can't tell you which item on that list is most important given what else you have going on today, or that the thing in slot 3 is actually blocked by the thing in slot 7, or that you're about to miss a commitment because you're focused on the wrong work. They're lists. Lists are fine. They're not thinking.
Claude can think. The question is what format to give it.
The format
The format is a plain text file — I call mine tasks.md — with three sections. Each item is a single line. The rules are minimal by design: the more structure, the more maintenance overhead.
tasks.md — the format
## NOW
- Final pass on inbox triage blog post (30 min)
- Review PR #14 — Marcus's comment on the auth flow
- Reply to Lisa re: invoice status
## NEXT
- Book 1 chapter 3 outline
- Update homepage with new convo block component
- Draft email sequence for Book 1 launch
## WAITING
- Marcus: contract signature (said EOW) — sent 2026-05-26
- Newsletter platform: DNS propagation — started 2026-05-27
- Lisa: invoice payment — followed up 2026-05-28
That's it. No priority numbers. No due dates on individual tasks (important deadlines go in a separate calendar). No tags or projects. Just the three states and one line per item.
WAITING is the most important state. Most task systems don't have it, so commitments you're waiting on from other people either get re-added to your NOW list (noise) or they disappear (dropped balls). The WAITING section captures them without cluttering your active list.
Working with Claude
Paste the task file into Claude and ask specific questions. The three most useful ones:
Claude working from the task file
Here's my task file. It's 2:15pm. I just finished the inbox triage post. What should I do for the next 90 minutes?
[paste of tasks.md]
Given the WAITING section, Marcus's contract signature is time-sensitive — "said EOW" with today being Wednesday. I'd prioritize a 10-minute check: do you need to follow up, or is it still on track?
After that, PR #14 review is a good 90-minute fit — it's collaborative work with a concrete output and it clears a NOW item completely.
I'd hold the chapter 3 outline for tomorrow morning when you have fresh focus. It's creative work and afternoon momentum doesn't usually serve it.
So: 10 min Marcus check → 80 min PR #14 review.
"The task file isn't the system. The thinking Claude does with it is the system. The file is just the input."
Maintenance
Update the file when things change. Move items from NEXT to NOW when you start them. Remove WAITING items when they resolve (or move them to NOW if the resolution created work). Archive completed items to a separate done section at the bottom of the file if you want a record — or just delete them.
The file should never be longer than one screen. If it grows past that, something's wrong — either you're tracking too many things at once, or you're not pruning completed items. Both are fixable.
Next step
The task file is one piece. Book 1 covers how it integrates with the morning open routine, inbox triage, and the end-of-day session close — so your task list stays accurate without extra effort.
Newsletter
No schedule. Just the things worth writing down — session memory, hooks, automations, multi-agent setups.