An AI-assisted weekly review works when the AI does the clerical layer — gathering what happened, grouping it, surfacing patterns, drafting the agenda — and you do the judging: what mattered, what slipped, what changes, per a split that productivity practice has documented for decades (the review itself, from David Allen's Getting Things Done onward) and that AI finally makes cheap, because the reason reviews die is the twenty minutes of assembly that precede the thinking. The AI spends the twenty minutes; you spend the judgment. Reverse the roles — let the model decide what your week meant — and you get fluent, confident nonsense about your own life.
RechargeMe publishes information, not advice. The workflow below uses documented capabilities of chatbots and note tools; no product testing, and your data terms apply to whatever you paste — route sensitive material accordingly.
What does the clerk role look like?
Five tasks, each a documented capability. Aggregation: paste the week's raw material — calendar exports, task-list states, meeting notes, journal scraps — and have the model produce a chronological and topical digest. Grouping: clusters of related activity (three separate threads that were one project; four small tasks that were one avoidance). Pattern-flagging: "meetings occupied 60 percent of working hours; deep-work blocks appeared twice." Data assembly: counts and tables — tasks completed versus created, which closes the loop most reviews skip. Agenda drafting: a review document with sections pre-filled from the above, ready for your annotations. Every output cites its input; nothing interpretive is asserted yet.
What does the judge role require?
Three decisions that don't delegate. Priorities: of everything that moved, what actually mattered — a judgment against goals the model doesn't hold. Honesty: the gap between the plan and the week is information; a model smoothing it into "great progress on several fronts" is worse than useless, so the agenda's first question — what did I intend that didn't happen, and what does that tell me — must be answered by you, plainly. And the adjustment: one or two structural changes for next week, not ten — reviews that generate ten resolutions change nothing, a failure mode every long-term practitioner documents. The model can hold the mirror; only you can decide what to do about the reflection.
| Review task | Clerk (AI) | Judge (you) |
|---|---|---|
| What happened | Digest and group the week | Correct the record |
| What it means | Flag patterns, no verdicts | Name what mattered |
| What slipped | List untouched intentions | Explain, honestly |
| What changes | Draft next week's structure | Pick one or two changes |
Related stories: An AI-friendly second brain that stays honest: capture, distill, cite · Batch your drafting: one AI session instead of forty scattered interruptions.
What is the actual session protocol?
Thirty to forty-five minutes, same slot weekly. Before: export or paste the week's raw material into your assistant with a saved review template — instructions fixing the digest format, the pattern-flagging rules ("list observations, not evaluations"), and the hard rule "do not editorialize about my performance." During: read the digest, correct factual misses — the model wasn't at your week, and its grouping will occasionally be wrong in ways only you can see. Then answer three questions in writing: what mattered, what slipped and why, what changes. After: one or two structural commitments into next week's calendar — blocked time, cancelled recurring meetings, a project promoted or demoted — because a review that ends without a calendar change is a diary entry.
What are the failure modes?
Documented temptations, in order of danger. Letting the model judge: its "key wins this week" framing is cheerleading, and accepting it replaces reflection with affirmation. Over-pasting: dumping sensitive personnel or health material into a consumer chatbot to save typing violates the first rule of tool selection — the review works fine with calendar and task data alone; keep the sensitive notes local. Review bloat: the template accreting sections until the hour becomes two and stops happening — cut sections that haven't changed a decision in a month. And the metric trap: counts the clerk computes easily (tasks done, messages sent) crowding out the judgments that are hard (was any of it the right work), which is the review dying of measurable-ness.
Why does this beat just using AI to "do productivity"?
Because the documented value of the weekly review was never information processing — it was forced confrontation with your own choices, weekly, in writing. AI makes the confrontation cheaper to reach, not less necessary. The practitioners who get the most from these tools use them to spend more time deciding and less time assembling — not to skip the deciding. That framing is also the honest hedge against the productivity industry's newest pitch: automation that reviews your week for you and tells you how it went. A summary of your week you never engaged with is a photograph of a gym.
FAQ
- What should I paste into the AI each week? Calendar, task-list states, and meeting notes — the mechanical record. Keep sensitive personnel and health notes out of consumer chat tools; the review works on the operational layer.
- How long should the review take? Thirty to forty-five minutes, of which assembly is now minutes; the thinking was always the point and still takes the time it takes.
- Can the AI tell me what my priorities should be? No — it can group, count, and flag patterns against goals you state. Choosing the goals is the job the review exists to protect.