Good morning.

You'll do about six jobs this week, and only one of them is the job you set out to do. The rest arrived because there's nobody else in the building.

Today I'll show you where each one goes, and give you the prompt that reads your own week and picks the first one to take off you.

—Sam

IN TODAY’S ISSUE 🤖

  • Four places any repeating job can end up

  • The prompt that audits your actual week

  • Six roles you're already doing yourself

  • What a workflow costs to build

  • The part of each job worth keeping

  • Where 69 hours a month end up

Four Places Work Can Go

When I’ve asked many of you what was getting in your way, the same answer kept coming back in different words:

One of you wrote that there are so many options and capabilities you don't even know where to start.

Someone else described the problem as "knowing which processes to automate first."

And a third person put their finger on the gap between theory and Tuesday morning: "I know a lot of the things I could do, but I have a hard time figuring out the steps to DO them."

That's a routing problem, more than anything else. You don't need another list of what's available. You need to know where a specific piece of your own work belongs.

Once you stop doing a repeating job by hand, it ends up in one of four places:

Where it goes

Use it when

What it costs to build

A prompt

The work turns up now and then

Nothing. You paste it.

A Skill

It happens the same way every time and needs your standard

An afternoon

An Agent

It has to start without you

A day, then a week of watching it

A workflow

Two of the above hand off to each other

Build it last

Ask yourself two questions to start figuring this out:

How often does it happen? If the answer is once in a while, a prompt is the whole solution.

Most people skip straight past this row and start designing an agent for something that happens twice a quarter, which is how you end up with a half-built system and a worse mood.

Does it have to start without you? If yes, it's an Agent. If you're content to trigger it, build the Skill, because it's cheaper to make and far easier to fix when it goes wrong.

A third question turns up later, once you've got two pieces running: Does the work cross a handoff, where one thing produces and another checks or sends?

That's a workflow, and it's the last thing you build rather than the first.

Now the part that matters if you're running this with one person or two:

You are the human here. The question is not necessarily who to hire. It's more about which sliver of each job needs your judgment, and how small you can make that sliver.

In the audit below, sixty-nine hours a month come out at fifty-three routed and sixteen kept, and almost all sixteen are time on the phone with a customer.

Start With Your Own Week

Before you look at anyone else's roles, find out what's in yours.

Paste this into whatever you use, with your real week attached:

Here is how my week actually went: [paste last week's calendar, your task list, and the subject lines from your sent folder]. I sell [what you sell] to [who buys it]. I run this business with [number] people, including me.

Go through this and pull out every piece of work that repeats, whether weekly, monthly, or every time a customer does something. Ignore the one-offs.

For each repeating piece of work, give me the work described in my own words, roughly how many hours a month it takes, how often it happens, and what breaks if it runs wrong for a month before I notice.

Route each one to exactly one of four places. A prompt, if it happens occasionally and I can paste it into a chat window when it comes up. A Skill, if it happens the same way every time and needs my standards applied, and I am content to trigger it myself. An Agent, if it has to start without me, watch something, or run on a schedule. A workflow, if it needs two or more of those chained with a handoff between them.

Assume I am not hiring anyone and cannot afford to. Where a piece of work genuinely needs a human, that human is me. Name the specific part that needs me, make that part as small as it honestly can be, and route everything around it.

Sort by hours per month, highest first. Then name the single one I should build first and say why that one before the others. If what I pasted does not contain enough repeating work to be worth building for, say so instead of stretching to fill the list.

I ran it against a real week: an independent marketing consultant, six retainer clients, around $180k a year, no staff.

Here's what came back:

Work

Hrs/mo

Goes to

Stays yours

Six weekly client check-ins

18

Skill preps each one, workflow sends the follow-up

The call itself

Content writing, posting, replies

10

Workflow: Skill writes, Agent schedules, Agent flags replies worth answering

One read before it publishes

Discovery call prep and follow-up

8

Skill

The call

Six monthly client reports

6

Agent

Nothing after the first month

Proposals

6

Skill

The number

QBR decks

6

Skill

The story you want it to tell

Invoicing and chasing overdue

4

Workflow: Agent watches dates, sends, escalates at 60 days

One phone call a quarter

Contractor draft review

4

Skill

Borderline calls only

Admin and expenses

3

Agent

Nothing

Competitor scan for one client

2

Agent

Deciding what's worth acting on

Contract renewal tracking

2

Agent

The negotiation

Sixty-nine hours a month, and about fifty-three of them route away.

Of the sixteen that stay, roughly twelve are the weekly check-in calls themselves and four are the discovery calls, which leaves the prep, the follow-ups, the reports, the decks, and the chasing on the other side of the line.

What survives is time on the phone with a customer, one read before something publishes, and a handful of judgment calls. Nobody was hired to take the rest.

It picked the monthly reports to build first. Six hours, judgment that writes down cleanly, a fixed date so it never waits on anyone to remember, and a bad month that's visible and cheap to fix. The invoice chasing came second, at a third of the hours, because the failure mode there is your own money sitting still.

The six roles below are the ones I see unfilled most often. Take whichever matches what your own audit put at the top.

Role #1: The Producer

Everything that turns one idea into the images, the video, and the cuts that a week of publishing eats.

Most operators write the idea and then lose an afternoon making it look like something.

The occasional version is a prompt. The version that runs every week is a Skill holding your visual standards, so you stop re-explaining your palette. The version that works ahead of your content calendar is an Agent.

Your part is the on-brand call, made once, written down, and then applied by everything downstream.

Start with the prompt:

Here's one thing I want to publish this week: [paste the idea, the post, or a section of something you've already written]. My brand looks like this: [describe your colours, whether you use photography or illustration, and two things you never want to see].

Give me a week of visual assets built from this one idea. For each, tell me the format and exact dimensions, what's in the frame, what text sits on it if any, and which platform it's for. Cover at least one static image, one short vertical video with a shot list, and one text-on-background quote card.

Write the image generation prompt for each static asset in full, ready to paste, and describe each video shot in one line with its duration. Keep every asset recognisably from the same idea, so the week reads as one campaign rather than seven unrelated posts. If any asset would need footage or a photograph I don't have, say so and give me the alternative that uses only what I can generate myself.

Role #2: The Analyst

The person who reads your own numbers and tells you what changed.

Almost nobody has this one filled, which is why most small businesses find out about a problem from a customer rather than from a dashboard.

This role is the clearest case for an Agent, because the work is a scheduled pull against a threshold and the judgment is entirely writable. Start with the prompt so you can see the shape of the answer you want, then build the Skill around that shape, then put it on a schedule. What stays yours is deciding what's worth acting on.

Attached is [the export: your payouts, your ad account, your store's order history, your email platform's last 90 days]. My business sells [what you sell]. The last time I looked at this properly was [when].

Tell me what changed. For every movement worth my attention, give me the number then, the number now, the percentage change, and your best read on the cause based only on what's in this file. Separate the movements that are seasonal or ordinary from the ones that are new.

Then give me the three things worth acting on, in order, with the specific action for each and roughly what it's worth if it works. Where the data can't tell you the cause, say what other file you'd need rather than guessing. Do not calculate anything you can't show me the working for.

Role #3: The Merchandiser

Listings, inventory levels, and the leak in ad spend that nobody checks.

If you sell products, this role is running whether you've staffed it or not, and it's usually running on whatever you wrote the week you launched.

Listing copy is a Skill, because it happens the same way every time and needs your voice. Stock levels and spend are Agents, because they have to watch when you're not looking. Pricing strategy stays with you and always will.

Here are [number] of my product listings: [paste titles, descriptions, and the current price of each]. I sell to [who buys, and what they're usually trying to accomplish].

For each listing, rewrite the title and the description. Lead with what the buyer gets rather than what the product is, use the words a buyer would type rather than the ones we use internally, and answer the single most common objection inside the description.

Give me the original and the rewrite side by side so I can see exactly what changed, plus one line on why. Where the problem is the product or the price rather than the words, tell me plainly instead of rewriting around it.

Role #4: The Support Desk

The job is knowing which messages need you, which ones your documentation already answers, and which ones are a customer starting to leave.

Triage is the piece worth building first, and it's an Agent, because a message arriving at 11pm shouldn't wait for you to sort it.

Reply drafting is a Skill. The angry ones and the refund conversations stay yours, permanently, and that's the right call rather than a limitation.

Here are the last [number] messages that came into support: [paste them]. Here's what I sell and the parts of it that break most often: [describe].

Sort every message into one of four buckets: answered by our existing documentation, needs a real answer from a person, an early sign of churn, and a sales opportunity in disguise. For each message tell me the bucket, the one line that put it there, and how urgent it is.

For the documentation bucket, name the specific page that answers it, and flag any question that came up more than once with nothing to point at, because that's the next thing I should write. For the churn bucket, tell me what the customer is actually worried about underneath the question they asked. Put anything you're unsure about in the person bucket rather than guessing.

Role #5: The Collections Clerk

Getting invoiced money into the account. Every operator I talk to under a certain size is doing this at 4pm on a Friday when they finally remember, which is exactly why it belongs to something that doesn't forget.

The whole role is an Agent watching dates with a Skill drafting at each stage.

The only human piece is the rare call where the relationship needs a voice, and you'll make maybe four of those a year.

Here's my accounts receivable: [paste the list with invoice number, customer, amount, date issued, and date due]. My payment terms are [terms]. My relationship with most of these customers is [ongoing or one-off].

Sort these by how likely I am to actually see the money rather than by how old they are. For each overdue invoice tell me the amount, the days past due, and which stage of chase it should be at.

Then write the message for each one. Stage one is light and assumes it was missed. Stage two names the terms and asks for a payment date. Stage three states what happens next and still leaves the relationship standing. Keep every message under 120 words and never use the word "just". Flag any invoice where the amount or the age means I should pick up the phone instead of sending anything.

Role #6: The Contract Reader

Reading what you're about to sign and telling you which three things to ask about.

Most people at this size sign whatever arrives, because a lawyer for every renewal costs more than the renewal.

Renewal and expiry watching is an Agent, and it's the version that saves you the most, since the expensive contract mistake is usually the one that rolled over while nobody was looking.

Redlining against your standard positions is a Skill. The negotiation is yours.

Here's an agreement I'm about to sign: [paste the full text]. I'm the [supplier or customer] here, the value is [amount], and the term is [length].

Read it and tell me what to look at before signing. Cover anything that governs how it ends, anything that creates an obligation with no ceiling, anything about who owns what gets made, and any clause where the two sides carry different risk for the same event.

For each item give me the clause number, what it says in plain language, what it means for me specifically, and whether it's standard for this kind of agreement or unusual. Rank them by what they'd cost me if things went wrong. Where a clause is fine, say nothing about it. This is a review to tell me what to ask about, so I'll take a lawyer's view on anything you flag as serious rather than acting on this alone.

When Two Of Them Chain

Workflow is a word that makes people picture a build they can't afford.

Here's the whole of one, using the Collections Clerk, and it's done in an afternoon:

An Agent reads your invoice list every morning and finds anything past its due date. It hands each overdue invoice to the Skill, which writes the message at whichever stage that invoice has reached.

The Agent sends it, logs the date, and moves the invoice to the next stage. When something crosses sixty days, the chain stops and puts it in front of you with a one-line summary and the customer's payment history.

Four moving parts, one handoff, and one simple escalation rule. The escalation rule makes the whole thing safe to leave running, because you've told it exactly where its authority ends.

Build the pieces in the order they'd fail:

  1. The Skill, so you can read what it writes before anything sends.

  2. The Agent that reads dates, running in report-only mode for a week.

  3. Connect the two.

  4. Last, let it send on its own.

Keep The List

Save this somewhere you'll see it, fill it in as you go, and add a row when you notice a seventh role you've been carrying.

# My Roles

| Role | Filled by | Still mine | Next |
| --- | --- | --- | --- |
| The Producer | | | |
| The Analyst | | | |
| The Merchandiser | | | |
| The Support Desk | | | |
| The Collections Clerk | | | |
| The Contract Reader | | | |

Rules I'm working to:
- A prompt is enough until the work happens weekly.
- A Skill is enough until the work has to start without me.
- I build the escalation rule before I let anything send.
- Nothing gets built twice. If it's on this list, it exists somewhere I can find it.

Last reviewed:

The column that matters is the third one. When you first fill this in, "still mine" will be long.

The job over the next few months is making each entry in it smaller and more specific, until what's left is only the work that needed you.

You were doing all six of these jobs before you read this, and nobody handed them to you on purpose.

What's new is that four of them can run without you sitting in front of them, and the part you keep gets smaller each time you look at it properly.

The Cortex build this month goes deeper on one of these, the part where customers slip away before anybody notices.

Talk soon,
Sam Woods
The Editor

.