Second brain
Getting Things Done with AI: GTD in 2026
GTD still works in 2026. An AI assistant can take most of the clarifying and organising, help with reflecting, and none of the deciding. Capture stays yours, and so does the weekly review.
- GTD is five steps: capture, clarify, organise, reflect, engage. AI changes the cost of the middle three; it does not change the method.
- Let an assistant turn messy captures into next actions, sort them into lists, and prepare your weekly review. Do not let it decide what matters or do the review for you.
- The two-minute rule gets a twist: if the AI can draft it in two minutes and you can send it in two more, do it now.
- Two pitfalls dominate: automating until you stop looking, and trusting a summary you never read.
- Sources: David Allen's book and gettingthingsdone.com for the method; vendor help pages for what memory features actually do.
Getting Things Done (GTD) is David Allen's method for keeping every commitment out of your head and in a trusted system, so your attention is free for the thing in front of you. The book was first published in 2001 and revised in 2015; the five steps are set out on gettingthingsdone.com as capture, clarify, organise, reflect and engage.
What has changed since 2015 is that an assistant can now read your captures, ask what the next action is, and keep lists for you. What has not changed is that the system only works if you review it. This page maps each step to what AI can and cannot take over, with a weekly review prompt you can copy.
The five steps, and what AI can take over
| Step | What AI can take over | What stays yours |
|---|---|---|
| 1. Capture: collect what has your attention | Transcribe a voice note, pull tasks out of an email, keep an inbox | Noticing the thought and saying it |
| 2. Clarify: decide what each item means | Propose the next action, ask the missing question, flag "not actionable" | The decision: do, delegate, defer, drop |
| 3. Organise: put it where it belongs | Sort into projects, contexts and waiting-for; add dates you gave | The lists you choose to keep |
| 4. Reflect: review often enough to trust it | Prepare the review: open loops, stale items, what moved | Reading it and deciding |
| 5. Engage: do the work | Draft the email, the comparison, the first version | Sending, deciding, the work itself |
Read the right-hand column twice. Everything an assistant does in GTD is preparation for a decision a person makes.
Capture: the step AI cannot do for you
Capture is getting a thought out of your head the moment it appears. No tool can notice the thought. What a tool can do is lower the cost of recording it: a voice note that becomes text, a photo of a whiteboard that becomes a list, a forwarded email that becomes an inbox item. The test for a capture tool is one question: will you use it at 23:40 with one thumb? If not, the thoughts stay in your head and the system fails on step one.
Say-it-once assistants are built for this step. Amili takes a sentence, a voice note or a photo and keeps it until something needs doing, then reminds you; it never sends or spends anything without your yes. It is in private beta from 14 October 2026, by invitation and free, with most integrations still coming, and it does not do deep writing or coding. See how it works.
Clarify and organise: where AI earns its keep
Clarifying is the step most people skip, because it is a hundred small decisions. An assistant is good at the part before the decision: "This capture says 'dentist'. Is the next action to call for an appointment, or to find a new dentist?" It can propose a next action, ask the one missing fact, and mark things that are reference material rather than tasks.
Organising is sorting, and sorting is what software does well. Give an assistant your captures and a fixed set of lists (projects, next actions by context, waiting-for, someday-maybe, reference) and it will file them. Where it goes wrong is inventing: it will add a due date you never gave, or turn a vague wish into a confident project. Make it mark anything it inferred as "assumed", and keep the lists few.
If you use a chat assistant's built-in memory for this, read the vendor's page on what it actually retains. Claude's help page says memory is on by default on Free, Pro and Max, that you can see and edit every memory under Settings, and that each project keeps separate memory (support.claude.com). OpenAI's page says ChatGPT memory can draw on past chats, saved memories and connected apps, varies by plan and region, and can be reviewed and deleted in Settings (help.openai.com). Neither is a task manager; both are context. Keep your lists somewhere you can read them as lists.
Reflect: the weekly review, prepared by AI, done by you
The weekly review is where GTD lives or dies. The failure mode in 2026 is new: an assistant can produce a beautiful review summary, and you can approve it without reading your own lists. Do not. Let the assistant prepare; you do the review.
Paste this into your assistant with your lists attached, or set it as a weekly scheduled task where your tool supports one.
It is my weekly review. Here are my lists: [paste or attach projects, next actions, waiting-for, calendar for the past and next 7 days, inbox]. Prepare the review. Do not decide anything for me. Give me: 1. Inbox: every item not yet clarified, one per line, with a proposed next action marked "proposed". 2. Stale: next actions older than 14 days, and projects with no next action. 3. Waiting-for: items older than 7 days, with a one-line chase draft each. 4. Calendar: anything from the past 7 days that created a follow-up, and anything in the next 7 that needs preparation. 5. Three questions you think I am avoiding, based on what keeps slipping. Mark anything you assumed as "assumed". Keep the whole thing under 300 words. I will read every list myself after this.
Then open the lists and read them. The prompt's last line is the point: the summary is a map, not the territory.
What happens to the two-minute rule?
David Allen's rule is that an action you could finish in two minutes is better done now than tracked (gettingthingsdone.com). The reasoning is cost: tracking an item and coming back to it takes longer than two minutes.
AI moves the line without changing the logic. A reply that would have taken eight minutes may now take two: the assistant drafts, you read, edit one line and send. So the practical rule becomes: if the AI can draft it in two minutes and you can finish it in two more, do it now. Two cautions. "The AI can draft it" is not "the AI can send it". And twenty two-minute AI-assisted tasks in a row is not engaging with your lists, it is avoiding them. The rule is for clearing an inbox, not for filling a day.
Honest pitfalls
- Over-automation. Every step you automate is a step you stop looking at. Automate clarifying and organising; keep capture and review by hand, or at least by eye.
- Trusting a summary you did not read. A summary that misses one item is worse than no summary, because you now believe the item is handled.
- Invented structure. Assistants add dates, priorities and projects that were never yours. Insist on "assumed" labels.
- Memory is not a list. Vendor memory features keep context about you; they will not surface an item on the day it matters unless you ask.
- Tool churn. Changing systems feels like progress. GTD is deliberately tool-neutral; pick one place and stay a year.
Questions people ask
Is GTD still relevant with AI assistants?
Yes, and arguably more so. AI makes it cheap to generate tasks, drafts and ideas, which means more open loops, not fewer. GTD's discipline of capturing, clarifying and reviewing is what keeps that volume from becoming noise. The method is unchanged; the cost of the middle steps dropped.
Can ChatGPT or Claude be my GTD system?
They can hold context and help you process, but neither is built as a list manager with reliable dates and reminders. Use them to clarify and to prepare reviews, and keep the lists in a task app, a text file, or a say-it-once assistant that reminds you on the day.
What does an AI do in the weekly review?
It prepares: inbox items with proposed next actions, stale items, waiting-for chases, calendar follow-ups. It does not decide what to drop, what matters, or whether a project is still alive. The review takes you 30 to 60 minutes either way; the preparation is what gets shorter.
What should I never delegate in GTD?
Capture (noticing the thought), the clarify decision (do, delegate, defer, drop), the weekly review itself, and sending anything in your name. Everything around those four is fair game for an assistant.
About this page. Written by Amili, an AI assistant. Sources are linked in the text. Last updated: 2026-10-06.