What to Automate With AI: Find Your First Job in a Week

Most people who try AI at work start in the wrong place. They open a chat window, ask for something clever, and then wonder why nothing in their week changed. The real question comes before any tool: what to automate with AI in the first place. If you pick a job that is rare, vague or high risk, even a very good tool will disappoint you.

This guide walks you from zero to a first job handed off. You will list your tasks, score them, pick one, and write the brief. It is the same path taught in Lesson 1.5, Find the jobs to hand to AI, which goes through it step by step. Read this first if you want the whole picture in one sitting.

Hand AI the task that repeats most often, follows a pattern you can describe, and has an output you can check in under a minute.

Why the choice of job matters more than the tool

A tool can only do what you can explain. If you cannot say what a finished job looks like, the tool has to guess, and guesses are where generic output comes from. The prompting guidance published by Anthropic makes the same point: treat the AI like a capable new employee who has no context on your norms, and explain precisely what you want.

So the work is mostly on your side. Choosing a job that has a clear shape makes everything after it easier. Choosing a fuzzy job means you spend the week fixing output instead of saving time.

Step 1: Write down the week you actually have

Do not start from what you think AI is good at. Start from your calendar, your sent folder and your task list for the last two weeks. Write every task as one short line. You are not judging yet, only listing.

Good places to look:

• Your sent email: which replies look almost the same?
• Your calendar: which meetings end with a follow-up you write by hand?
• Your messages: which questions do customers ask again and again?
• Your files: which documents do you rebuild from an old copy each time?
• Your week-end: what do you always do on Friday afternoon because it never got done earlier?

Aim for 25 to 40 lines. If you reach ten and stop, you are listing only the big tasks. The small repeating ones are where hand-offs usually pay back.

Step 2: Mark the five that repeat

Read your list and circle anything that happened at least weekly, or that you know you will do again next month. Keep the five that feel most similar each time. A task you do once a quarter can wait, because even a perfect hand-off saves you very little.

Step 3: Score each one on three questions

For each of your five tasks, give a score from 1 to 3 on how often it repeats, how long it takes, and how easy it is to hand off. Higher is better on all three.

Add the three numbers. A total of 7 to 9 is a strong first candidate. The third question is the one people skip, and it matters most. A task you do daily but that depends on a delicate decision about a person or a refund scores low on handoff, however often it comes up.

Step 4: Check the risk before you pick

Two tasks with the same score are not equally safe to start with. Before choosing, ask:

• If the output is wrong and I miss it, what happens?
• Does the task involve private customer data, money or anything legal?
• Can I read the result in under a minute and know whether it is right?

Pick a job where a mistake is cheap and visible. A first draft of a weekly update to your own team is a better start than a message that goes straight to a customer. AI tools can make mistakes, and the vendors say so on their own pages, so keep yourself in the loop on anything that leaves your business.

Step 5: Pick the one job

Take the highest score that also passed the risk check. Write it as a single sentence in this form: "Every [time], I [do this], using [these inputs], and the result is [this output]." If you cannot finish that sentence, the task is too fuzzy. Go back to your next candidate.

Step 6: Write the brief that hands it over

The brief is where the hand-off actually happens. Keep it plain. It needs five things:

1. The jobOne sentence on what you want made and why it exists.
2. The inputsWhat it will be given each time, such as a list of notes, a spreadsheet or an email thread.
3. What good looks likePaste one real example you are happy with, or describe length, tone and layout in plain words.
4. What to avoidClaims you cannot back, phrases you dislike, topics that are off limits.
5. Questions firstAsk it to check with you about anything unclear before it starts.

Anthropic's guidance lists examples as one of the most reliable ways to steer output format, tone and structure, and suggests a few varied ones. If you have three past examples of the finished task, paste them all. For more on this habit, see How to Brief AI Like a Senior Hire.

Step 7: Run it once and judge it

Run the job on real material from last week, where you already know the right answer. Compare the result with what you did by hand. Look at three things: is it correct, does it sound like you, and did it save you effort even after you checked it? If the answer is no on any of them, change the brief and run it again. Do not change the job yet.

If you use a tool that works on files on your own computer, such as Claude Code, which its documentation describes as a tool that can edit files and run commands, the same brief applies. The only change is that the output can be a saved file rather than text in a chat.

What to do after the first job works

Once the first job runs well for two or three weeks, go back to your scored list and pick the next one. Each new job gets easier because you now have a brief template and a sense of what to check. Keep the list somewhere you will see it, and add new repeating tasks as you notice them.

Over time, your question changes from "what can AI do" to "what in my week is still done by hand and does not need to be". That is a better question, and it keeps you in control of the work.

Questions people ask

How many tasks should I list?

Between 25 and 40 is a good range. Fewer than that usually means you left out the small repeating tasks, which are often the best candidates.

What if my best candidate involves customer data?

Start with a lower risk job first, such as internal notes or drafts only you will see. Check the privacy terms of any tool before putting customer information into it.

Should I automate the task I hate most?

Only if it also scores well on frequency and handoff. Disliking a task does not make it easy to explain to AI.

How do I know the first job worked?

Compare it to last week's work done by hand. If you can check it in a minute and it needs only light edits, it is working.

Where to go next

If you want a guided version of everything above, Lesson 1.5 covers it in order: list your week, score each task, pick the first job and write the brief that hands it over. You can find it here: Find the jobs to hand to AI. Pair it with the Find framework if you want a repeatable weekly routine.