The Five Keeps: Which Decisions to Keep for Yourself

There's a document in my business that tries to predict how I'll decide things.

My agents keep it up to date, and it's built from decisions I've actually made: times I approved something, times I sent something back, and times I removed an approval step I didn't need.

When one of my agents runs into a decision it thinks I might want to weigh in on, it reads that document first and asks itself whether I'd really want to be bothered.

At the top of the document is a one-line summary of how I make decisions. Boiled down, it says this: protect the premise, the promise, the relationship, and anything that can't be undone.

Delegate the reversible work, and demand proof that it's finished.

Add money to that list, and you have every kind of decision I keep for myself. Everything else, my AI team can decide.

That's a short list, and I want it to be short. When agents decide and act without waiting on you, you need systems that let you hand that much over and still trust the result.

And trust starts with knowing which few decisions stay on your desk.

Below, I'll walk you through the five kinds of decisions I keep, why each one stays with the owner, and how to sort your own decisions into the two piles.

Can It Be Undone?

Before we get to the five, let's start with the question that does most of the sorting: Can this be undone?

Most of the work in a small business can be.

A blog post can be edited. A page layout can be changed back. A social post can be deleted. A new email sequence can be paused.

A report can be redone. A design choice can be revised next week.

If an agent gets one of these wrong, you fix it, and the cost is a little time.

When work can be undone, it belongs with the team, even when customers will see it. That's the whole idea behind a launch-then-evaluate habit.

The team does the work, checks it, puts it live, and fixes or rolls it back if it misses.

Some work can't be undone, though, or can only be undone at a high cost.

A wire transfer is gone once it's sent. A contract is binding once it's signed.

A promise to a customer is made the moment they read it, even if you take the email back later. A deleted account with years of data may never come back.

And a few decisions can technically be reversed but shape everything that comes after them. If you change who your business serves, you can change it back, but you'll have spent months building the wrong thing in the meantime.

Those are the decisions I keep. I've grouped them into five categories, and I call them the Five Keeps: the Premise, the Promise, the Relationship, the Money, and Irreversible risk.

Keep 1: The Premise

The premise is your strategic direction. It answers questions like: What business are we in? Who do we serve? What are we building this year?

What are we going to stop doing?

AI agents are very good at how. Give them a clear objective, a clear audience, and a clear offer, and they can work out the steps, build the thing, and check their own work.

They shouldn't invent the objective. A team that sets its own direction is no longer your team.

Here's how that plays out in my business. If the question is which layout, workflow, or wording best serves a page we've already agreed on, the team decides.

If the question would change who the page is for, what it's selling, or why it exists, it comes to me.

The premise also works in the other direction. Once I've settled it, I don't want to be asked again.

When I approve the main article for a topic, the team turns it into social posts, emails, and video scripts without coming back to me for each one.

Asking me to approve every piece made from something I already approved just slows everything down. Approving the premise once is enough.

When I asked my team to work through a packaging decision for one of our brands, I asked them to weigh the tradeoffs among themselves and bring me one strong recommendation instead of a menu of options.

A menu hands the thinking back to me. A recommendation lets me do the one thing that's mine, which is to say yes, no, or "not that direction."

Keep 2: The Promise

The promise is what your business tells the world it will do. It includes your core offer, your guarantee, your brand, and any specific commitment to a specific customer, like a price, a scope, or a delivery date.

Your team can say a lot of things on your behalf. In my business, agents write project updates, answer client questions, and keep customers informed every day.

They can't make a new promise, though. An agent can tell a client where their project stands.

It can't tell that client the project will be finished by the 15th unless that date has already been agreed. It can't offer a discount, add a feature to the scope, or describe a guarantee we don't offer.

Your brand belongs here, too. Inside an established brand, I'm happy to let the team make creative choices: spacing, images, colors from the existing palette, the wording of a headline.

Once something has been approved, it becomes the pattern for the work around it, and the team extends that pattern without asking.

But a new visual identity, a new tagline, or a new way of describing what we do is a change to the promise, so it stays with me.

Why keep this one? Because the promise is your name. Customers forgive a typo on a web page.

They have a much harder time forgiving a commitment you didn't keep, and they won't care that an AI agent made it.

Keep 3: The Relationship

The third keep covers your key relationships and the messages that have to come from you.

In my business, AI agents communicate with clients all the time. They do it openly, as members of the team, under their own names.

That works well, and I don't approve those messages one at a time. But every final message written under my name, I send myself.

My agents can research it, organize my thoughts, and prepare everything I need, but if it's signed "Steve," Steve sends it.

That's the line I draw for identity. A message that claims to be from me has to actually be from me.

The relationship keep goes beyond signatures, though. Some relationships are important enough that you'll want to stay closer to them for a season.

I have one client relationship where everything the team produces still comes to me before it goes out, because of the nature of that account.

That's a choice I made for that relationship alone. It doesn't apply to the rest of my clients.

Major reputation questions belong here, too.

If a customer is angry enough to post about you in public, if a partner is about to walk away, or if something has gone wrong in a way people will talk about, you want to be the one deciding how to respond.

Your team can gather the facts and recommend a response, and then you make the call and, usually, do the talking.

And then there are the messages that carry weight precisely because they come from you: the launch announcement, the story of why you started, the hard conversation with an employee.

You write or record those first. Then you let AI multiply them across every channel you have.

Keep 4: The Money

The fourth keep is money: pricing, spending, and payments.

In my business, I approve final payment links, purchases, refunds, new subscriptions, pricing changes, and contracts. My agents can do almost all the work around those decisions.

They can research a tool, compare the plans, estimate what a price change would do to our revenue, and put together the page that explains it.

The decision itself stays with me.

This next rule is especially important. My agents are expected to use the accounts and subscriptions I already pay for.

For example, we generate video through a service I've already chosen and funded. If that account runs out of credits, the agent stops and asks me to refill it.

It doesn't get to switch to a different service with a separate bill just because that would let it finish the job today.

An AI agent that's trying hard to complete a task may look for another way to get it done, and sometimes the other way costs money you never agreed to spend.

The rule is simple: an agent can spend your time freely, because time spent on reversible work comes back to you in finished tasks. It can't spend your money without you.

Keep 5: Irreversible Risk

The last keep is anything you can't take back: legal, security, and anything else that can't be undone.

That includes things like:

• contracts, terms of service, privacy policies, and other legal language
• statements about your security or compliance, like telling a client their data is protected in a certain way
• changes to domain names and DNS, which can take your website and email offline
• creating, deleting, or changing the permissions on user accounts
• billing systems and anything that charges a customer
• any change that doesn't have a tested way back

In my business, routine website maintenance runs on its own, with a fresh backup and a written rollback plan before every update.

Look at what that same policy leaves out, though. The moment a change would touch DNS, billing, email delivery, payment behavior, or user accounts, it stops and comes to me.

So does any change where a backup or rollback isn't available.

The work is the same kind of work. The only difference is whether there's a way back.

I keep it to one sentence: If there's no tested way back, it doesn't launch without me.

Everything Else Goes to the Team

Here's what's left after the Five Keeps, and it's almost everything:

• doing the work inside a direction you've already set
• routine updates to customers and clients
• creative and design choices inside your established brand
• research, analysis, and recommendations
• reversible launches with good checks and a way to roll back
• scheduling, follow-up, reporting, and administration
• internal tools, documents, and processes
• turning one approved idea into many pieces of content

Keeping a decision means you make the call, even when you don't do the work. On a pricing change, your team might do 95 percent of the work: the research, the numbers, the new page, and the email to customers.

Your part might take ten minutes. That's still a keep, and it's still a good use of AI.

The list can also change over time. For a while, I kept the final say on our video editing.

At the end of this past August, I handed routine video work to the team, with an independent quality check built in.

Making that change was itself a decision only I could make, and once I made it, the work moved to the other pile.

As your team proves itself in an area, you may find that something you thought was a keep is really just a habit.

What about your own thinking? Your reading, writing, and time with customers aren't decisions, so they don't go on this list.

They're how you get good at the five keeps. An owner who stops thinking will make bad premise decisions, and no AI team can fix that.

Try This: Sort Your Last 20 Decisions

Look back at decisions you've actually made and sort them into two piles.

Step 1: Find 20 decisions. Look through your sent email, your calendar, your text messages, and your task list from the last two weeks.

Write down 20 decisions you made. Include the small ones: "approved the new blog layout," "told the client we'd have it by Friday," "renewed the scheduling software," "picked the photo for the newsletter."

Step 2: Ask the questions. For each decision, ask:

1. Can it be undone easily if it's wrong?
2. Does it set or change the premise (direction, audience, offer)?
3. Does it make or change a promise (price, scope, date, guarantee, brand)?
4. Does it involve a key relationship or a message under my name?
5. Does it commit money?
6. Does it carry irreversible risk (legal, security, accounts, no way back)?

Step 3: Sort them.

Decision	Undo? (Y/N)	Premise	Promise	Relationship	Money	Irreversible	Delegate or Keep

If the answer to question 1 is yes and every other column is blank, mark it Delegate. If any of the five columns is checked, mark it Keep.

Step 4: Write your keeps down. Write a short list of your Five Keeps in your own words, with an example or two of each from your sorted decisions.

Step 5: Give the list to your team. Put it in the standing instructions for your chief of staff.

Tell it something like this: "Decide everything that isn't on this list. Bring me only what is."

Step 6: Hand off three. Pick three decisions from your Delegate pile that you're still making yourself out of habit, and tell your team they own them from now on.

Most of your 20 decisions will probably end up in the Delegate pile, and that's the point of the exercise.