AI Weekly Review

A Pro feature that reads your current state and writes a narrative weekly review — summary, per-task suggestions, and commitments extracted from your notes.

How it works

When you click Generate with AI at the top of your Weekly Review, MLW sends a structured snapshot to Anthropic's Claude and gets back:

1. 3-paragraph narrative summary — what you shipped, what drifted, one suggested focus for next week. Grounded in your actual data (specific project names, dates, task counts).
2. Per-item suggestions — for Overdue, Stale Next Actions, and Someday items, Claude proposes an action (complete / drop / next / someday / rewrite) with a one-line rationale. Each appears as an "AI:" pill next to the task's normal action dropdown. Click the pill to accept, × to dismiss.
3. Notes → Tasks extraction — commitments Claude finds in your recent notes that you probably should turn into tasks. "Call Linda Thursday" in a Wednesday daily page, for instance. Each shows with the quoted excerpt and an Add as Next / Add as Inbox button.
4. Area observations — one-sentence observations when Claude notices an Area went quiet or showed unusual concentration.

What gets sent

• Your active tasks (title, status, project, area, dates, modified timestamp)
• Your active projects (name, outcome, task count, most recent activity)
• Recent notes from the last 7-14 days (title + body preview, truncated)
• Your Areas of Focus and their activity counts

Completed tasks go in too, so Claude can write the "what you shipped" section.

What does NOT get sent

• Your email address or identity
• Completed/dropped tasks older than the review window
• Notes from outside the review window
• Any data from other users

Privacy and consent

First time you click Generate, you'll see a consent modal explaining what's sent. Accepting writes a flag to your profile. You can revoke anytime in Settings → AI Review.

Anthropic does not retain your request content beyond serving the response.

Limits

• 31 generations per month per Pro user. Resets on the 1st of each month.
• 5 generations per hour.
• Regenerating within 15 minutes returns your existing review; after that, a fresh one is generated.

Tips

• Let Claude be wrong sometimes. Its suggestions are grounded but not infallible. Dismiss the ones that feel off. The point is to accelerate your review, not replace your judgment.
• Run it weekly, not daily. More frequent = less signal. Claude works best with a week of change to narrate.
• Read the summary before the per-item pills. The narrative primes you for the tactical decisions. Flipping straight to pills is like reading the score before the game.

---

Next: Search →