GTD with AI: automating the next action

David Allen's Getting Things Done runs on five steps: capture, clarify, organize, reflect, engage. Almost everyone who tries GTD gets the first four working for a while. Almost everyone loses the fifth one, the weekly review, within a couple of months. That's not a discipline problem. It's a design problem AI can actually fix.

GTD's core idea is that your mind is for having ideas, not holding them, so every commitment gets captured the moment it appears and moved into a trusted external system. From there it's clarified into a concrete next action, organized into the right list, reviewed on a schedule, and engaged with when the time is right. It's one of the most widely adopted personal productivity systems ever written, and the reason it works when it works is that the lists stay current. The reason it stops working is almost always the same: the review that keeps them current gets skipped.

The five steps, and where each one strains

David Allen has called the weekly review the critical success factor in making the whole method stick, and it's easy to see why: skip Reflect for a few weeks and Engage stops being trustworthy, because you no longer believe your lists reflect reality. That's the exact same collapse every method in this space hits eventually, and it's worth reading the fuller argument on why second brain systems fail, and the one fix.

What AI actually automates here

Capture and clarify are where AI helps first and most obviously. A conversation, a meeting transcript, or an email thread almost always contains commitments buried in the middle of something else, and an assistant that's actually present for those conversations can pull out "next action: send the revised proposal by Friday" without you having to notice it yourself and write it down. That's capture and clarify happening as a side effect of work you were already doing, not a separate task you have to remember to do. The same read-and-extract instinct is what makes PARA automatable too, just with next actions instead of buckets.

Organize follows the same pattern: once a next action exists, filing it under the right project or context is mechanical, and it's exactly the kind of judgment an assistant with the right context can make reliably. The step that changes the most, though, is Reflect. Instead of a dedicated weekly session where you manually reconstruct what's current, the system can already be current, because capture, clarify, and organize happened continuously all week. The review becomes a genuine check-in rather than the moment you rebuild the whole list from scratch. That's the exact shift covered on the weekly review template, run by AI, and it plays out the same way across every method on the PKM methods comparison page.

A concrete place this shows up

GTD's "waiting for" list, the commitments other people owe you, is one of the first things that goes stale in a manual system, because nobody enjoys re-checking it. An AI-run version of GTD can watch for replies and flag the ones still outstanding automatically, which is the same problem tackled from a different angle in AI follow-up tracking, no CRM. If you're arriving here from outside a technical role and wondering whether any of this applies to how you actually work, Claude Code for non-developers covers who this is realistically for.