A Worked Week: Picking What to Automate With AI, Day by Day

People often want proof that choosing what to automate with AI works outside a demo. We have no real numbers to give you here, and we will not invent any. What we can give you is a plan for a single week, shown through an invented business, so you can see what each day involves and run the same week on your own work.

The business below is made up. Nothing in it is a result. If you want the lesson behind it, Lesson 1.5 covers each step: Find the jobs to hand to AI.

In one week you can list your tasks, score them, pick one safe job, write a brief and run it once on real material, which is enough to know whether to keep going.

Meet the example business

Say Dana runs a three-person home cleaning company. She does bookings, schedules, quotes and customer follow-ups herself, mostly from her phone and laptop. She is busy, a bit sceptical of AI, and has tried a chat tool a few times without it sticking.

Day 1: Fetch the week

Dana opens her sent email, calendar and messages for the last two weeks and writes one line per task. She does not judge, only lists. She ends up with 31 lines, including quote replies, confirmation texts, rota messages to staff, review requests, invoice reminders, answers about what products she uses, rescheduling notes and a weekly summary for her business partner.

What to copy: the habit of working from records, not memory.

Day 2: Mark what repeats

Dana circles anything weekly or more. Eleven tasks qualify. She crosses out those that need her judgement every time, such as complaint replies, and keeps seven.

What to copy: drop the tasks that need you personally before you score anything.

Day 3: Score the seven

She gives each task 1 to 3 for frequency, duration and ease of handoff, and adds them up.

Three tasks tie at 7. Dana now has a shortlist, which is more useful than a feeling.

Day 4: Check the risk and choose

Quote replies involve prices and promises, so she wants to read every one before it goes, and she is not ready for that. Booking confirmations go to customers but follow a fixed pattern. The rota message goes only to her staff and she can check it in seconds. She chooses the rota message.

What to copy: break ties by choosing the job where a mistake is cheapest and easiest to see.

Day 5: Write the brief

Dana writes the five parts. Her notes are plain words, and she follows the idea in Anthropic's prompting guidance that the AI is like a new employee who lacks context, so she explains it all.

1. The jobWrite the weekly rota message for my cleaning team, so everyone knows who is where and when.
2. The inputsA list of jobs with address, time and who is assigned, pasted from my schedule.
3. What good looks likeLast week's message pasted in, which is short, friendly and lists jobs by day.
4. What to avoidNo promises about pay or hours, no changes to the schedule I gave it.
5. Questions firstAsk if any job is missing a time or a name before writing.

She keeps the brief in a note so she can reuse it.

Day 6: Run it on last week

She feeds in last week's schedule, where she already knows the right message. She reads the result with the original beside it and checks three things. Is every job present and correct? Does it sound like her? Did she need to change much? She notes where it went wrong, perhaps a swapped time or a stiff tone, and changes the brief, not the output.

What to copy: test on old work where you know the answer.

Day 7: Decide and write it down

Dana decides to use the brief for the next two rotas, reading each before sending. She writes down what to watch: any wrong time or name, and how long checking takes. She sets a date to review in two weeks. She does not touch the next job yet.

What the example does not tell you

It does not tell you how much time Dana saves, because the example is invented. It does not tell you that AI will get the rota right every time. AI tools can make mistakes, and the vendors say so on their own pages, which is why Dana keeps reading each message before it goes.

What it does show is the order of work. List, mark, score, check the risk, brief, test, decide. Each day has one small output, and nothing depends on trusting the tool early.

Run your own week

Here is a version you can copy.

1. Day 1List every task from the last two weeks, from records. Aim for 25 to 40 lines.
2. Day 2Circle anything weekly or more. Remove tasks that need your judgement every time.
3. Day 3Score frequency, duration and handoff ease from 1 to 3 and add up.
4. Day 4Check the risk and choose one job, breaking ties by cheapest mistake.
5. Day 5Write the five part brief with a real example.
6. Day 6Run it on last week's material and compare with what you did.
7. Day 7Decide, write down what to watch and set a review date.

If you use a tool that works with files on your computer, such as Claude Code, which its documentation describes as able to edit files and run commands, the same week applies. Only the output form changes.

Things that might go wrong in your week

• Your list is too short. Go back to your sent folder before scoring.
• Everything scores low. That is information. Try the "second time" list for a month.
• The first run is poor. Fix the brief, especially the example, and run again before judging the job.
• You are tempted to add a second job. Wait until the first has run for two or three weeks.

The routine behind this is covered in the Find framework, and the brief is explained in How to Brief AI Like a Senior Hire.

Questions people ask

Is Dana a real customer?

No. She and her business are invented to illustrate the method. No results are claimed.

Can I really do this in seven days?

The steps are small enough to fit around work, usually less than an hour a day. Take longer if you need to.

What if I score everything and nothing stands out?

Pick the safest of the top scorers. A clear first job beats a perfect one.

Should I send the AI output without reading it?

Not while you are learning. Read each one, and keep reading anything that goes to a customer.

Where to go next

Start your own Day 1 today. When you want a guided version, Lesson 1.5 walks through the whole week in order: Find the jobs to hand to AI.