How I Built a System That Tells Me What to Automate Next, Every Single Week

The boring tasks eating your week are invisible to you.

You sit down to automate something this week.

You scroll back through your week, pick whatever felt slow, and burn a Saturday building a routine for it. A few weeks later it’s running, sort of. You can’t tell if it gave you back any time. The boring tasks you didn’t see are still eating your days. The thing you automated wasn’t even in the top five.

You can’t see what to build. You’re inside your own week, and the patterns that matter are invisible from there.

Here’s the fix. A small audit system that reads your AI sessions, finds the patterns you keep missing, and hands you a ranked list of what to build next.

By the end of this article you’ll know what the system becomes over four weeks, and you’ll have the full build to make it real by Sunday night.

Why You Pick the Wrong Thing

Most automation projects get picked from memory. You scan back over the week, try to remember what felt slow, and pick from there.

Our memory is terrible at this.

Memory is terrible at this.

It remembers the dramatic stuff. The launch email that took two hours. The client crisis on Wednesday. It forgets the boring stuff. The seven times you reformatted the same metric for a Substack post. The fourteen times you opened your dashboard to check the same number.

Boring is WHERE the time goes.

The audit reads your actual sessions instead of your memory. It logs what you did and clusters what repeats. Then it hands you a ranked list with the patterns ordered by time-saved-per-week.

That’s the whole idea. Now what it becomes.

What the Audit Actually Does

Four jobs.

It reads your AI session history. Chats, prompts, outputs, and the runs you abandoned halfway through.

It clusters by pattern. What repeats. What takes longest. What you keep redoing because the first version was off.

It compares this week against last week. What’s new. What’s gotten worse. What you stopped doing without noticing.

It hands you a ranked list. Top five candidates for automation, ordered by time-saved-per-week, with the specific session files that show each pattern.

Run it once a week. Sunday night works. You wake up Monday to a one-page report telling you what to build that week.

The first run will surprise you. The top item is usually obvious in retrospect.

The items at #4 and #5 are where the real value sits. Those are the patterns you didn’t know existed.

Week 1: The Loud Stuff

The first run catches the obvious tasks. The ones you’d recall on a long flight without trying. The launch email that ate three hours. The recurring research request you keep redoing.

These come back as the top items. Time-saved-per-week, ordered, anchored to the actual session files.

You go after the easy wins first. A few hours back in the next week.

That part the audit catches because the patterns are loud. Anyone with a notebook could find them. The compounding starts when the loud stuff is gone.

Week 2: The Layer Underneath

Week 1 catches the loud stuff. The tasks you do a lot. You automate them and free up a few hours.

Week 2 reads a different week. The loud stuff is gone, because you automated it.

Now the audit sees what was sitting underneath.

This is usually meta-work. The prep before the writing. The research before the cold email. The image sourcing before the publish.

The unglamorous setup steps nobody thinks of as tasks because they don’t feel like work, they feel like getting ready.

They are the most expensive minutes in your week. You’ve never measured them.

Build those automations. Free up another batch of hours.

Week 3: The Chains

By the third week the audit has enough history to spot something the human brain almost never sees on its own.

Workflow chains.

A chain is when you have several tasks that are actually one process pretending to be many.

The audit notices that every Tuesday you do four things in sequence:

1. Pull this week’s open rates from Substack

2. Drop them in a tracking doc

3. Write a two-line summary for yourself

4. Queue a thank-you DM to the highest-engaging new subscriber

Four separate tasks in your head. One process in reality.

The recommendation will be to collapse them. One routine that runs all four in order and drops the result in your inbox before coffee.

Task-level thinking misses chains. They only show up when something is reading across a few weeks of sessions at once.

Week 4: Your Blind Spots

By the fourth week, the audit starts doing something most people don’t expect.

It flags missing work.

Tasks you SHOULD be doing on a recurring basis but aren’t. Things like:

• You haven’t checked your free-to-paid conversion rate in 18 days

• You haven’t followed up on the four replies that came in two weeks ago

• You haven’t logged any new content ideas in 9 days, even though you logged 23 of them in March

• You haven’t run your own audit in 11 days and the report is overdue

The audit catches your drift. Because it can read everything you’ve done, it can also see what you’ve stopped doing.

This is the moment automation thinking flips.

You went in trying to save time. You came out with something that watches the operation and tells you what’s slipping.

What You Just Saw

That’s what the audit becomes over four weeks. The reader who builds it walks into Monday already knowing what to ship.

The full build is below.

Both prompts (the session logger and the audit itself), the scheduling trigger, the output landing zone, and the master-task-list template that makes Week 4 work the way it just sounded like it could.

The free side of The AI Handbook ships frameworks like this every week. The paid side ($15/month) ships the implementation. If you got something out of this concept, the vault has dozens more builds, prompts, and operator systems lined up. One system that gives you a few hours back pays for the year on its own.

---

---

How to Build It

Five pieces. They’re all small.

1. The session logger. Most AI tools keep chat history in the cloud. The audit needs files. Skip the copy-paste step entirely and have Claude write the session log itself.

At the end of every work block in Cowork or Claude Code, drop this one line:

Save a clean summary of this session to /AI-Sessions/[today's date].md.
Include what I worked on, the prompts I used, what I built, what I 
abandoned, and how long the session ran.

Claude has file access in both tools. It writes the markdown. You move on. The /AI-Sessions folder is the audit’s reading material.

2. The audit prompt. This is the one that runs weekly. Drop it into a saved skill or a routine:

Read the session files in /AI-Sessions from the last 7 days. 
Identify tasks that appear 3+ times this week. 
For each pattern, estimate the total time spent. 
Flag any pattern where I built something twice instead of automating once. 

Output a ranked list with the top 5 candidates for automation, 
ordered by time-saved-per-week. Include the specific files or 
sessions that show each pattern. 

Then read /master-task-list.md. End with a section called "Drift" 
that lists tasks from the master list I've stopped doing this week, 
plus any recurring patterns from past weeks that have gone silent.

3. The master-task-list. This is the file that makes Week 4 actually work. Without it the audit can only catch drift from your past sessions. With it, the audit knows what you SHOULD be doing on a recurring basis.

Save this template at /master-task-list.md:

# Master Task List

## Weekly
- Publish 2 newsletter issues
- Check free-to-paid conversion rate
- Follow up on reader replies
- Log new content ideas
- Run the weekly audit

## Bi-weekly
- Review and clean Notion content board
- Update the operator-profile.md file
- Audit subscriber growth and source breakdown

## Monthly
- Deep audit of pipeline and offers
- Update voice-rules.md if anything has shifted
- Review and retire automations that aren't pulling weight

Edit the items to match your actual recurring duties. The audit reads this list every Sunday and flags anything you’ve gone quiet on.

4. A scheduled trigger. A Claude Routine, a cron job, a calendar reminder, or a Zapier-scheduled webhook. Whatever fits your stack. Sunday night, before the new week starts.

5. An output destination. A Notion page that overwrites each Sunday, a self-email, or a pinned doc in Drive. Whatever you’ll actually open on Monday morning.

That’s the whole system.

Run the audit once. The first week’s output will look obvious. Something like: you spent four hours drafting Substack outlines from scratch this week, build an outline drafter.

Build the outline drafter and save the four hours. The compounding starts next week.

What This Actually Becomes

The audit is a feedback loop. Build it once and it stays useful.

Set up an automation and walk away. Six months later, half are quietly broken and the other half are automating things you don’t even need anymore. Nobody is checking on them.

The audit catches that. It tells you what to build next, what to retire, what’s quietly broken, and what you should have built three weeks ago.

The audit itself saves you zero time. It points you at the NEXT ten hours of building that will.

That’s the gap between someone who automates a few tasks and walks away versus someone who builds a real operating system around their week.

Pick a Sunday. Build the audit. The week you can’t see right now is the same week the audit starts handing back to you, in pieces, every Monday morning.

Ryan