A long to-do list creates a strange kind of fake productivity: everything looks important, so you keep switching between tasks instead of finishing one meaningful piece of work.
AI can help here, but not by deciding your priorities for you. The useful job is narrower: turn a messy list into a small set of explicit choices, define what “done” means, expose dependencies, and build one realistic 90-minute sprint you can actually execute.
In roughly 40 minutes of setup and review, you can create four useful outputs: a priority filter, a finish-line card, a 90-minute sprint plan, and a post-sprint learning loop.
Capability check — August 16, 2026: OpenAI says ChatGPT Projects are available across free and paid plans and can keep chats, files, and project instructions together. OpenAI’s current Free Tier FAQ also lists web search, file/image uploads, and data analysis, subject to usage limits. Google currently documents NotebookLM standard access at up to 50 sources per notebook and 50 chat queries per day. None of those features is required for this workflow: a normal text chat, a notes app, and a timer are enough.
Before you start: give AI only the work context it needs
You do not need to upload your entire inbox, project drive, calendar, customer database, or company workspace just to plan one focus session.
A safer input is a short redacted list containing:
- task name;
- deadline if it is real and explicit;
- who is waiting for it;
- dependency or blocker;
- rough effort based on your own judgment;
- consequence of not doing it today;
- whether another person must review or approve it.
Remove passwords, private customer data, confidential financial information, employee-sensitive material, unreleased strategy, and anything your organisation does not permit you to process in that AI service.
The model should organise your evidence, not invent a hidden priority system.
1. Cut a 15-item list down to three real candidates
Outcome: Three tasks that are plausible candidates for the next focus block, plus a clear reason the rest can wait.
Best fit: Developers, analysts, designers, marketers, managers, researchers, students doing project work, and anyone whose day starts with a mixed list of small and large tasks.
Inputs and tools: Your redacted task list and a basic text-capable AI assistant. A spreadsheet or plain note works fine as the source.
Steps
- Paste the task list with the factual fields above.
- Ask the AI to score only what the data supports: deadline pressure, dependency impact, reversibility, and whether the task can make meaningful progress in 90 minutes.
- Separate urgent because of evidence from emotionally loud.
- Force the model to show missing information instead of silently guessing.
- Pick the final top three yourself.
Useful interaction pattern:
Rank these tasks only from the information I supplied. Use four factors: explicit deadline, number of other tasks/people blocked, consequence of delay, and whether meaningful progress is possible in one 90-minute session. Show your reasoning in one sentence per task. Write UNKNOWN where the list does not contain enough information. Do not infer business priority, client importance, or urgency from task wording alone.
Time and cost: About 10 minutes. A free text chat is enough.
Privacy, accuracy, and copyright limits: AI cannot know the political, commercial, or human importance of a task unless you state it. A “small” task may be critical because of context you did not include. Treat the ranking as a decision aid, not an instruction. Do not upload private documents merely to improve the ranking.
2. Turn the chosen task into a finish-line card
Outcome: A tiny definition of what the session will produce, what it will not attempt, and how you will know when to stop.
Best fit: Work that expands while you are doing it: research, coding, writing, design, planning, debugging, analysis, and presentations.
Inputs and tools: One chosen task plus any short, approved context needed to define the deliverable.
The goal is to avoid a vague target such as work on dashboard or prepare proposal. A focus sprint needs a visible artifact or decision.
Steps
- State the task in one sentence.
- Add the recipient or user if there is one.
- Ask for the smallest useful output you can finish or materially advance in 90 minutes.
- Define three acceptance checks.
- Add an explicit not today list.
- Review the card before starting the timer.
A good finish-line card might look like:
| Field | Example |
|---|---|
| Sprint result | Draft the API migration plan through risk review |
| Done when | Scope listed, dependencies mapped, top 3 risks written |
| Not today | Implementation, ticket creation, stakeholder presentation |
| Blocker | Need confirmation on legacy auth ownership |
Useful interaction pattern:
Turn this task into a 90-minute finish-line card. Give me: one concrete output, three observable acceptance checks, a NOT TODAY list, one likely blocker, and the exact point where I should stop instead of expanding scope. Do not invent requirements or deadlines. If the task is too large for meaningful progress in 90 minutes, propose the smallest coherent slice and explain what was excluded.
Time and cost: About 10 minutes.
Privacy, accuracy, and copyright limits: The model can make a scope sound neat while misunderstanding what your team actually needs. Check the finish line against the real request, ticket, brief, or assignment. If you use third-party material, keep excerpts minimal and make sure you are allowed to process them.
3. Build a 90-minute sprint with only two focus blocks
Outcome: A realistic sequence that protects one task from constant switching.
Best fit: Work that benefits from concentration but still needs a short reset midway through.
Inputs and tools: The finish-line card, a timer, and optionally your calendar or task app. AI is used to sequence the work; it does not need access to your calendar.
Use a simple 90-minute shape:
| Time | Activity |
|---|---|
| 0–5 min | Open only the files/tools required; write the finish line where you can see it |
| 5–40 min | Focus block 1: create the first coherent chunk |
| 40–45 min | Short break; no inbox or social feed |
| 45–80 min | Focus block 2: finish, test, or tighten the result |
| 80–90 min | Verify acceptance checks, record next step, close the workspace |
The exact minutes are not sacred. The important rule is that the sprint has one result and very few transitions.
Steps
- Give the AI the finish-line card.
- Ask it to split the work into no more than four execution steps.
- Put the highest-uncertainty step first when possible.
- Define what to do if you hit a blocker: record it, make one bounded attempt, then switch to the next useful subtask rather than opening five new searches.
- Start the timer and stop planning.
Useful interaction pattern:
Convert this finish-line card into a 90-minute execution plan using 5 minutes setup, 35 minutes focus, 5 minutes break, 35 minutes focus, and 10 minutes verification/closeout. Use no more than four work steps. Put the riskiest unknown early. Add one fallback action if I hit a blocker. Do not add meetings, research branches, tools, or deliverables that are outside the finish-line card.
Time and cost: About 10 minutes to create the plan. The execution itself is 90 minutes. A free timer and notes app are enough.
Privacy, accuracy, and copyright limits: AI is bad at estimating your personal execution speed from a task title. If the plan contains six subtasks that each take half an hour, it is not a plan—it is a wish list. Cut it down. Do not connect work calendars or internal tools when manual input is enough.
4. Run a 10-minute post-sprint review before creating the next plan
Outcome: A short record of what actually consumed time, what blocked progress, and what should change in the next sprint.
Best fit: Anyone who repeatedly underestimates tasks, gets derailed by the same dependencies, or finishes focus sessions without knowing what to do next.
Inputs and tools: Your finish-line card, the actual result, and three quick notes: what finished, what did not, and why.
Steps
- Mark each acceptance check as
DONE,PARTIAL, orNOT DONE. - Record where the plan diverged from reality.
- Separate bad estimate, missing information, external blocker, and scope expansion.
- Ask the AI for one change to the next sprint—not ten productivity tips.
- Write the next action in a form another person could understand.
Useful interaction pattern:
Compare my finish-line card with what actually happened. Classify each miss as bad estimate, missing information, external blocker, scope expansion, or distraction. Do not moralise about productivity. Identify the single planning change most likely to improve the next 90-minute sprint, and write one concrete next action. Keep assumptions clearly labelled.
Time and cost: About 10 minutes.
Privacy, accuracy, and copyright limits: A post-sprint review is not a performance evaluation. Do not use a language model to judge employee effort, capability, or motivation from one block of work. Keep the review about process evidence you can observe.
The 40-minute setup-and-review loop
| Activity | Time |
|---|---|
| Cut the list to three candidates | 10 minutes |
| Define one finish-line card | 10 minutes |
| Build the 90-minute sprint | 10 minutes |
| Review the result afterward | 10 minutes |
Then run the actual 90-minute focus sprint separately.
This separation matters. If you spend the whole session asking AI how to be productive, the tool has become the distraction.
A free or low-cost route
You do not need a dedicated AI productivity subscription.
- ChatGPT Free: OpenAI currently documents web search, file/image uploads, and data analysis on the Free tier, subject to separate usage limits. For this workflow, plain text is usually enough.
- ChatGPT Projects: available across free and paid plans and useful if you want the same project instructions and approved reference files available over several work sessions. Free projects currently have lower file limits than paid tiers.
- NotebookLM standard access: Google currently lists up to 50 sources per notebook and 50 chats per day. It can be useful when the sprint depends on a small source pack and you want answers tied closely to those materials.
- Plain notes + timer: the lowest-friction option. Keep your task list and finish-line card in a note, use AI only for restructuring, and close the chat when the sprint begins.
Keep AI outside the focus block when it is not needed
A useful rule is to distinguish planning assistance from continuous assistance.
Before the sprint, AI can help compress ambiguity. During the sprint, open it only when the task itself genuinely requires it—for example, analysing supplied data, checking a draft, or explaining a technical error.
Do not keep asking for a better plan every time the work feels uncomfortable. That turns planning into avoidance.
A simple fallback sentence is enough:
I am blocked on [specific point]. Based only on the context below, give me the smallest next diagnostic step I can complete in 10 minutes. Do not redesign the project or add new goals.
Five things AI should not decide for you
- which colleague, customer, or project “matters most” when you did not supply that context;
- whether a deadline is negotiable;
- whether confidential work is safe to upload;
- whether your performance was good or bad;
- whether a task should quietly expand beyond the agreed scope.
The useful output is not an AI-approved day. It is one deliberate result, a bounded session, and better evidence for the next decision.
Sources
Checked August 16, 2026: