Back to blog

Claude Routine or AI Employee? How to Tell Which One You Need

by Samantha 15 September 202610 min read

The first things I set up in Claude were routines. A morning brief that's ready before I'm at my desk, a knowledge base updater that runs every night, an inbox assistant that runs every hour and a meeting notes routine that checks for new notes every 30 minutes. They're still some of the most useful things in my business.

But none of them could take a piece of work from me. The morning brief could tell me a blog post was due. It couldn't write it, and it couldn't show it to me when it was done.

That's the difference I want to explain, because it's the one that confuses people most. People set up a routine and think, "I've got an AI employee now."

A routine is just a reminder. It wakes AI up at a set time to run a skill or check something.

An AI employee is built around the workflow. It picks up work when you give it a task, does it, and hands it back to you for review.

I use both, every day. The COO employee handles everything ad hoc, and anything that has to happen every single day becomes a routine, delegated to an employee.

Side by side

A routine An AI employee
What starts it The clock A task you give it
What it works from The same prompt every time Your brief on the task
What you get back A report, or a job done the same way as yesterday A finished task, sitting with you for review
Who decides what it works on The schedule You

If you only remember one line from this post, remember the first row. A routine starts because it's 6am. An employee starts because you gave it something to do.

What a routine is

A routine is a scheduled wake-up. It's three things: a time, a prompt, and usually a skill. At that time, Claude wakes up, reads the prompt, uses the skill to do the job and goes back to sleep.

These are the routines that run in my business:

  • The Morning Brief, every morning at 6. It reads my ClickUp and the status page for each of my clients and tells me what today is.
  • The Knowledge Base Updater, every night at 22:37. It reads what happened that day and writes it into the right client's log and status page.
  • The Inbox Assistant, every hour. It sorts my email and drafts the replies that are waiting on me.
  • Meeting Notes, every 30 minutes. It files the notes from my calls in the client's folder and turns the actions into tasks.

Every one of those is the same job, at the same time, every time. That's what a routine is good at.

What a routine can't do

A routine doesn't know about new work you give it mid-week. It runs the prompt it was given when you set it up. If I decide on a Wednesday that I need a proposal written, my morning brief will happily remind me about it on Thursday, but nothing in that routine is going to write the proposal.

It also has no "whose turn" and no review step. A routine finishes its job and reports. There's no point where it stops, hands the work to me, waits for me to say "change this", and then picks it back up with my notes.

That's fine for a morning brief. It's a problem for anything that needs your brief going in and your yes coming out.

What makes an AI employee

An AI employee has four parts.

  1. A skill. This is how to do the job: who it is, what it knows, the steps, what it never does. If you haven't built one, How to Build a Claude Skill walks you through it.
  2. A workflow. A task, a field that says which employee owns it, and statuses that say whose turn it is. Mine is ClickUp, with a field called AI EMPLOYEE and statuses like "w/ agent to do" and "w/ samantha for review". The full setup is in How to Build the Workflow for Your AI Employees.
  3. A way to pick up work. Something has to look for tasks that are waiting. Mine is a small script called the dispatcher that checks ClickUp every 10 minutes for any task with an AI EMPLOYEE set and a "w/ agent" status, and starts Claude on each one.
  4. A handback to you. When it's done, it comments on the task with what it did, and moves it to "w/ samantha for review". It lands in my inbox in ClickUp, and I read it, approve it or tell it what to change.

The skill is the same kind of thing a routine uses. What makes it an employee is the other three parts.

When to use which

Same job at the same time every day means a routine. Different work that arrives whenever means an employee.

When I'm not sure, I answer these five questions.

ROUTINE OR AI EMPLOYEE?

1. Is it the same job every time?
   Yes → routine. No → employee.

2. Does it need to happen at a set time, whatever else is going on?
   Yes → routine. No → employee.

3. Does it need a brief from you before it can start?
   Yes → employee. No → routine.

4. Does it need your review before it's finished?
   Yes → employee. No → routine.

5. Does the work arrive whenever, rather than on a schedule?
   Yes → employee. No → routine.

Mostly routine: set up a routine.
Mostly employee: build an employee, & let a routine wake it up.

That last line is the combo, and it's how most of my employees run. I'll show you that in Step 2.

What you need before you start

  • The Claude desktop app on a paid plan, with the Claude Code tab. Routines are set up from there.
  • A skill for the job. The routine and the employee both use one.
  • Your AI brain. A routine starts every run with no memory of the last one, so everything it needs to know has to be written down somewhere Claude can read it. If you haven't set that up, start with The Co-Founder Setup.
  • ClickUp, or whichever task manager you already use, for Step 2. The steps say ClickUp; swap in yours.
  • Your Mac awake with Claude open while you want it to run. If the laptop's shut, it runs when you next open it.

You don't need a developer or any code for either step.

Step 1 - Set up a routine

Open the Claude desktop app, click the Claude Code tab and start a new chat. Click the + at the bottom of the chat and choose your AI brain folder. Then paste this, filled in.

I want to set up a routine. Every weekday at [TIME], use my [SKILL NAME]
skill to [THE JOB], & save what you write to [FOLDER] in my AI brain.

If one thing it needs can't be reached, say so at the top & carry on with
the rest. Never invent anything.

Show me the routine before you save it, & tell me when it'll first run.
Then run it once now so I can see what it does.

Claude writes the routine, saves it and tells you when it'll first run. You'll find it in the Scheduled section of the Claude app, which is where you pause it or change the time later.

What you'll see straight away is the first run: the job done once, in the chat, so you can check it before tomorrow morning. If something's wrong, tell Claude in the same chat and it updates the routine.

The line about carrying on matters more than it looks. Without it, a routine that can't reach one of your tools gives you nothing instead of most of something. There's more on the different types in How to Build a Claude Routine.

Step 2 - Upgrade it into an employee that checks your ClickUp queue

This is the upgrade. The routine stops doing one fixed job and starts checking a queue for the work you've given it.

First, connect Claude to ClickUp.

  1. In the Claude desktop app, open Settings, then Connectors.
  2. Find ClickUp and click Connect.
  3. Sign in on ClickUp's own screen and allow access.

You sign in on ClickUp's screen, never in a chat, so there's no password or key to copy anywhere.

Then, in the same chat as Step 1, paste this.

Now I want to turn this into an AI employee that works from my ClickUp
queue instead of doing the same job every time.

First, walk me through adding what I need in ClickUp, one step at a time:
a dropdown field called AI EMPLOYEE with one option, [EMPLOYEE NAME], & two
statuses on my [LIST NAME] list, "w/ agent" & "w/ [MY NAME] for review".

Then change the routine. Every 30 minutes from [START] to [END], check
ClickUp for tasks where AI EMPLOYEE is [EMPLOYEE NAME] & the status is
"w/ agent". If there are none, stop & say "queue empty".

For each task, oldest first & no more than two a run: read the brief in
the task description, do the job using my [SKILL NAME] skill, save the
output to [FOLDER], comment on the task with what you did & a link, then
set the status to "w/ [MY NAME] for review". Never touch any other task.

Claude tells you exactly where to click in ClickUp to add the field and the statuses, then updates the routine and tells you when it'll next run.

To test it, create one task on that list. Write the brief in the description, set AI EMPLOYEE to your employee, and set the status to "w/ agent". Within half an hour you should see a comment on the task saying what it did, with a link to the output, and the task sitting at "w/ [your name] for review".

If you want changes, comment on the task with exactly what to change and set it back to "w/ agent". The next run picks it up with your notes.

This is the combo. The routine is the alarm clock, and the employee is the one who does the job. My Blog Manager works this way. Every half hour from 7am to 9pm, a small check looks at my Blog Posts list in ClickUp. If a post is waiting to be written, redrafted or published, Claude starts and does it. If nothing's waiting, it goes straight back to sleep.

The catch, and the fix

An employee only knows about the work you put in the task. If I explain what I want in a chat window, or in a voice note that never makes it into ClickUp, the employee never sees it. It isn't in the queue, so as far as the employee's concerned it doesn't exist.

The fix is one habit. The brief goes in the task description, every time, and anything you'd add later goes in a comment on that task. The workflow post has a task brief template you can copy, so the employee can work it without coming back to ask: How to Build the Workflow for Your AI Employees.

The second catch is that both of these run on your Mac. If your laptop is shut at the time a routine is due, it runs when you next open it. For a morning brief that's fine. For an employee you want working through the afternoon while you're out, it means leaving the Mac awake with Claude open.

What this looks like as a business

Most business owners you support don't need twenty AI employees. They need someone who can look at their week and say which jobs are routines and which ones need an employee, and then set up the right one.

That's the conversation you're paid for. You're not paid for the time it takes to set up a routine. You're paid for knowing that the Monday report is a routine and the client proposals are an employee.

Start with one

Pick one job you do every week and answer the five questions. If it's a routine, set it up with Step 1 this week and leave it alone for five days. If it's an employee, set up the field and the two statuses, give it one task, and read what comes back.

If you want to see what an employee looks like when it runs a whole business, I wrote up the first one I'd build: my AI COO.

If you haven't built a skill yet, that's where every routine and every employee starts: start here.

this is the free stuff — imagine what we'd build together

Ready to make your business AI-enabled?

Consulting and one-to-one coaching for founders and freelancers who want AI actually working inside their business — not just talked about.