5 Mistakes in Choosing What to Automate With AI, and Fixes

Hand-offs to AI usually fail for dull reasons. The tool is fine, but the job was a poor choice, the brief was thin or nobody checked the result. If you are deciding what to automate with AI, these five mistakes are the ones to avoid, each with a fix you can apply in minutes.

These are patterns that follow from how the work goes, not findings from a study. The full method for choosing a job is taught in Lesson 1.5: Find the jobs to hand to AI.

Most wasted time comes from choosing a vague, rare or risky job, and from skipping the brief and the check.

Mistake 1: Choosing the task you hate, not the task that fits

It feels natural to hand over whatever you dislike most. But dislike has nothing to do with whether a task can be explained to AI. A task you hate may be full of judgement calls, exceptions and mood.

Why it costs time: you spend a week fixing output for a job that was never easy to hand off.

The fix: score each candidate from 1 to 3 on how often it repeats, how long it takes and how easy it is to hand off, then add the numbers. Let the total decide. If your most disliked task scores low, leave it with you, or split out the patterned part.

Mistake 2: Picking a task that is too vague

"Handle my marketing" or "deal with customer messages" is not a job, it is an area. AI has no way to know what finished looks like, so it produces something generic.

Why it costs time: vague jobs generate vague results, and you cannot tell what to fix in the brief.

The fix: write the job as one sentence in this form: "Every [time] I [do this] using [these inputs] and produce [this output]." For example: "Every Monday I turn my meeting notes into a short list of action items with names and dates." If the sentence will not come out, the task is not ready.

Mistake 3: Writing a one line brief

You know your business well, so a short instruction feels complete to you. To an AI, it is missing everything. Anthropic's prompting guidance describes the AI as being like a capable new employee who lacks context on your norms and workflows, and says the more precisely you explain what you want, the better the result.

Why it costs time: thin briefs lead to rounds of "shorter, warmer, not like that."

The fix: use five parts in every brief.

1. The jobOne sentence on what and why.
2. The inputsWhat it receives each time.
3. What good looks likeA real example you like, pasted in. The same guidance lists examples among the most reliable ways to steer format and tone.
4. What to avoidClaims, phrases and topics that are off limits.
5. Questions firstAsk it to check anything unclear before writing.

For a longer walk through, read How to Brief AI Like a Senior Hire.

Mistake 4: Skipping the check

A confident tone is not the same as being right. AI tools can make mistakes, and the vendors say so on their own pages. If you hand over a job and stop reading the output, an error will eventually reach a customer.

Why it costs time: one wrong price or name can take far longer to repair than the task ever saved.

The fix: choose jobs you can check in about a minute, and set a rule for yourself on what you always read before it leaves the business. For anything involving money, legal wording or health, rely on official sources and a qualified person, not on a draft.

Run the job first on last week's material where you already know the right answer. That is a safe way to see how often it is wrong before you rely on it.

Mistake 5: Handing over too much at once

After a good first result, it is tempting to hand over five jobs in a week. Then something goes wrong and you cannot tell which brief, which input or which job is at fault.

Why it costs time: problems stack up and you spend your time debugging instead of working.

The fix: run one job for two or three weeks until it is stable and you know how long checking takes. Only then pick the next one from your scored list.

If you use a tool that works with files on your own computer, such as Claude Code, which its documentation describes as able to edit files and run commands, the same rule applies. A tool that can do more makes the check more important, not less.

A quick self audit

Look at any hand-off you already have and ask:

Each "no" points to one of the five mistakes above. Fix the highest one first.

A short illustration

This is invented, to show how the mistakes stack. Say a freelance designer wants to automate "client communication" because she dislikes it (mistake 1). She writes "reply to clients nicely" as her brief (mistakes 2 and 3), pastes in nothing else, and sends the results without reading (mistake 4). She also starts a second job for invoices in the same week (mistake 5). Nothing here is a real case, and no outcome is claimed. The fix is the same boring routine: score, define, brief, check, one at a time.

What good looks like instead

A calmer version of the same designer's week: she lists her tasks, finds that "send the weekly project status note" scores 8, writes it as one sentence, briefs the AI with her best past note, reads every draft, and leaves the other jobs alone for now. For the routine behind this, see the Find framework.

Questions people ask

Which mistake is the most common?

We have no data on that. In practice the first two, wrong job and vague job, come before everything else, so fix them first.

Can a good brief make up for a bad job choice?

Not fully. A brief cannot give structure to a task that has none.

How do I know a job is checkable?

If you can look at the output and say right or wrong within a minute or two, it is checkable.

What if I have already made one of these mistakes?

Nothing is locked in. Fix the brief or swap the job, and run it again on past material.

Where to go next

Pick one mistake from the list and fix it this week. If you want a guided path that avoids all five, Lesson 1.5 covers it step by step: Find the jobs to hand to AI.