AI Guides › Playbooks
By Nigel Guy · 7 min read
Most people start the day by rereading a to-do list that has quietly grown to thirty items, then doing whichever one shouts loudest. Asking an AI to "organise my day" usually makes it worse: you get the same thirty items back, neatly grouped under headings, and it feels like progress. It is sorting theatre. Nothing has been cut, nothing has been ranked against what you are actually trying to achieve, and the guilt items are still there tomorrow.
The rule: a daily plan is only useful if it forces a choice. Give the model your goal as well as your list, and make it hand back three things to do and a named pile of things to stop.
A list records obligations; it does not rank them. Every item carries equal visual weight, so urgency wins by default, and urgency is mostly other people's priorities. What a list lacks is the third input: the one or two outcomes you are trying to move this month. Without that, neither you nor a model has anything to rank against, so "reply to Dave" and "finish the pricing page" look identical.
The mechanism below adds that missing input and fixes the output to four boxes.
Every morning, the Project returns the same card, in the same order:
| Box | What goes in it | The test |
|---|---|---|
| 1. Do today (max three) | Tasks that move your stated goal, or that cost something real if they slip again | Can you name the goal it serves, or the cost of delay? |
| 2. Can wait | Things that feel urgent but lose nothing by moving a day | What actually breaks if this is tomorrow? |
| 3. Hand off | Work that does not need you specifically | Could someone else (or a template) do 80% of it? |
| 4. Drop | Items kept out of habit or guilt | Would you add this today if it weren't already there? |
Box 4 is the one that earns its keep. A plan that never drops anything is just a longer list.
You need a Claude account. At time of writing, Anthropic's help centre says Projects are available to all users, including free accounts, with free users limited to five projects. Pro is billed at $20 a month in the US (about £15 at time of writing) with local pricing elsewhere; check the price shown at checkout for your UK cost, as it depends on VAT and how you are billed. Free is fine to start.
Step 1 — Create the Project. Go to claude.ai/projects (or open Projects from the sidebar), select + New Project, and name it something like "Morning triage". Claude does not read the name or description, so they are just for you.
Step 2 — Add your standing priorities as Project knowledge. Use the + on the Project knowledge panel to add a short text file called "What I'm moving forward". Two or three outcomes, each with a deadline if there is one. This matters because, per Anthropic's own help article, context is not shared between chats in a Project unless it sits in the knowledge base. Yesterday's chat is invisible to today's.
Step 3 — Paste the instructions below via Set project instructions, then Save instructions.
Step 4 — Each morning, open a new chat in the Project and paste today's calendar and list. Keep it to one chat per day so old days do not bleed into the new one.
Optional: connect your calendar. Anthropic offers a Google Calendar connector (set up under Customize > Connectors, then switched on per chat via + > Connectors). Be aware that Anthropic's own pages disagree at time of writing: the connector docs say it is read-only and available on Pro, Max, Team and Enterprise, while a help-centre article describes it as available to all users and able to create events. Check what your account shows. Pasting your calendar works on any plan and lets you see exactly what the model saw.
Fill in [YOUR_ROLE] and [WORKING_HOURS]; everything else arrives in your morning paste.
You are acting as my chief of staff for daily planning. I am [YOUR_ROLE], and I normally have [WORKING_HOURS] of usable time on a working day.
Context: The Project knowledge file "What I'm moving forward" lists my current outcomes. Each morning I will paste three things: [TODAYS_CALENDAR], [TODAYS_TASK_LIST], and optionally [ANYTHING_CHANGED] (new deadlines, things that slipped yesterday).
Goal: Turn that into a decision, not a tidier version of the same list. A good result leaves me knowing exactly what to start on and what I have permission to ignore.
Before sorting:
1. If the knowledge file is missing, empty or clearly out of date, or I have not pasted a task list, stop and ask me for it. Do not infer my priorities.
2. Work out how many free hours my calendar actually leaves. If the three tasks you pick will not fit, say so.
Then return this card, in this order, using these exact headings:
DO TODAY (no more than three): each task, the outcome it serves or the cost of delaying it again, and a rough time estimate.
CAN WAIT: tasks that can move to tomorrow, each with one line on why nothing breaks.
HAND OFF: tasks that do not need me personally, with who or what could take them if I have told you; otherwise just flag them.
DROP: tasks that look like habit or guilt rather than current value. Be direct; do not hedge these.
ONE QUESTION: the single thing you would need from me to plan tomorrow better, or "none".
Rules:
- Use only what I have pasted or what is in Project knowledge. Never add tasks, people, deadlines or meetings I have not mentioned.
- Urgency alone does not earn a place in DO TODAY.
- Every task I pasted must appear in exactly one box.
- Keep the whole card under 250 words.
Self-check before replying: count the DO TODAY items (three or fewer), confirm every pasted task is placed exactly once, and confirm nothing in the card was invented.
The "every task appears exactly once" line is deliberate: it stops the model quietly losing items it would rather not judge.
Hypothetical: you run a two-person design studio. Your knowledge file says "Launch the new pricing page by Friday" and "Land one retainer client this quarter". This morning you paste four meetings (three free hours left) and eleven tasks.
A sound card would put the pricing-page copy, the follow-up to a warm retainer lead, and the invoice that is already overdue in Do today; move the portfolio tidy-up to Can wait; flag social scheduling for Hand off; and put "research new project-management tools", which has sat on the list for weeks, in Drop. It would also note that three free hours is tight for all three tasks. If the card instead put "reply to newsletter" in Do today, that is your signal the knowledge file is too vague.
The trap is a stale priorities file. The card is only as good as the goal it ranks against, and Claude will faithfully rank today's list against last month's goals. Update "What I'm moving forward" whenever an outcome is finished or abandoned, and glance at it every Monday.
The second trap is pasting fifty items. The model will place them all, but the card becomes unreadable. Do a thirty-second cull yourself first.
It will not know about anything you did not paste, including the email that arrived at 7am or the conversation where your manager changed the plan. It will not remember yesterday unless you tell it. It cannot judge politics: "drop" is a suggestion, and some guilt items are guilt items because a person is waiting on them. And even with a calendar connector, it will not move meetings for you on the read-only version.