---
title: "How do you build a weekly digest that turns agent output into visible work?"
published: 2026-10-18T13:05:00+03:00
updated: 2026-10-23
lang: en
canonical: https://ofershap.github.io/en/posts/weekly-agent-digest/
original: https://ofershap.github.io/posts/weekly-agent-digest/
---

# How do you build a weekly digest that turns agent output into visible work?

Set a fixed format with three questions - what came in, what turned into work, what got stuck - and run the daily output through it once a week, at a fixed time. The digest is not another report. It's the decision point where output becomes action or gets dropped.

## In depth

### The problem with daily output

An agent that works every day produces a pile: summaries, reminders, suggestions, links. Nobody reads a daily pile for long. What doesn't get used gets forgotten, and what gets forgotten gets missed. One week without review, and the output becomes background noise.

### The format: three questions, not a table of contents

What came in? The list of items the agent produced this week, unedited. The goal is coverage, not readability.

What turned into work? For every item that became an action - a task, a message sent, a decision - one line. This is the only metric that matters: output that turned into motion.

What got stuck? Items that came back week after week with no motion. A stuck item gets an explicit decision: push it forward, drop it, or change the way it arrives.

Example: one week brought fourteen items. Five turned into work: two messages sent, two drafts queued, and one document updated. Three items got stuck for a second consecutive week, and the decision was to take them off the daily list instead of dragging them along again.

### Rules that keep the digest useful

A fixed time. Same hour every week, regardless of load. A digest postponed twice is dead.

A decision criterion for every stuck item. If an item hasn't moved in two weeks, it leaves or changes. No third option of "we'll decide later".

A small number. If the list is too long for a ten-minute review, the problem is the daily output, not the digest. Shrink the source, not the review.

### What this gives you in practice

The digest turns the agent from an output machine into a work tool. Instead of remembering to check what happened, there's a fixed point where you see what came in, what turned into work, and what got stuck. The template I work with is in my public skills library, and you can install it as is.

## From my experience

In that same post I told how, after the recordings stopped for a week, I added a weekly digest: a fixed point where you see what came in, what turned into work, and what got stuck. Here is the full format and the rules that keep it useful.

## Frequently asked questions

### What's the difference between a digest and a daily summary?

A daily summary describes what happened. A weekly digest is a decision point: what turned into work, what got stuck, and what you decide about each stuck item.

### How long should the weekly review take?

About ten minutes. If it needs more, the daily output is too noisy and worth shrinking at the source.

### What do you do with an item stuck two weeks in a row?

Decide: push it now, drop it, or change the form it arrives in. No fourth option of quiet postponement.

## About the author

Ofer Shapira builds and leads AI in engineering teams, and writes about agents, decisions, and workflows you can verify.

Last updated: 2026-10-23