Part 3: Your first AI job is ready for a real test.

Part 3 of 4 | Build and test your first AI job

You picked a job in Part 1 and signed off on its job card in Part 2.

That job card is your saved game plan.

What kicks the job off, what the AI should produce, how it checks its work, and where it still needs you to weigh in.

Now it's time to put that card to work on one real case and see how it holds up.

By the end, you'll have one finished result you can actually use, plus a Skill saved for next time.

This first run might turn up something your job card missed.

That's fine. Coz you can still fix it now, before it becomes the reusable version.

Head back to your AI Brain Starter project.

You may use same chat you used in Part 1 and 2 if you still have it open, or a new chat in the same project if not.

Either way, it will used your saved job card to run this job.

For example, you might have the these inputs for the following jobs:

💡 Check your job card, next to “I will use”

The AI already tells you what input it needs to do the job.

If the info's already in the business files you uploaded in Part 1, just tell the AI where to find it.

Otherwise, paste it in or attach it.

💡 Check your job card, next to “Start it by saying”

This is the one-line prompt that starts this job (and your future runs of this job).

On this first run, the Starter uses your approved job card to build the reusable Skill, then gets on with the job.

Every run after this, that same short request just uses that saved Skill.

The AI should hand you:

If the AI doesn’t do the job, check Optional Help below for a few possible fixes.

Correct everything that needs correcting now, while you still make the business calls and give the final approval.

Read the whole result before deciding if it's good enough.

Compare it against the approved job card and the info you used for this case.

Look hardest at prices, deadlines, named owners, and recommendations. Any of these can turn into a commitment if they're wrong.

Ask yourself:

The result might include a “What you decide” section.

This covers calls the AI couldn't make from the saved instructions or the info it had.

You can bundle all your decisions and feedback into one message if that's easier:

You don't have to accept whatever the AI suggests.

Ask it to walk through what each option would actually mean if you need more context.

Tell the AI what's wrong, missing, or unclear. It should return the full result again, with your fix folded in.

After each revision, check it made the change you asked for without touching anything else.

Keep going until the whole result matches the job card you approved.

Before it closes the first job, the AI may ask whether a correction should be kept for future runs.

Some decisions belong to just this case.

While others can be useful direction for every time you run this job.

For instance:

Before a correction becomes a permanent rule, read exactly how the AI plans to word it.

Make sure it's only keeping what should actually repeat.

The AI will show you what it updated.

Job-specific rules go into the Skill, while broader working preferences can live in project memory.

Once you're happy with the result, tell AI you accept it:

AI should now save the result you accepted, the Skill, and the information and checks used for this run.

Congrats!

From this point forward, you now have a working one-line prompt that will run this job for future cases.

Part 4 will test the Skill on another case and give the job a repeatable route.

That is where you'll see whether it can handle new input without carrying over details from this run.

See you there.

If it hasn't named an actual blocker, just say:

If AI says it can't finish, read the reason before nudging it along.

It should tell you exactly what's missing or which information disagrees, and which part of the work that affects.

You can then supply the missing detail or settle the conflict.

If the problem affects only part of the job, you can ask:

Keep the unfinished part clearly flagged, and review the whole result again once that gap's closed.

If you decide to make the the job smaller instead, update and re-approve the job card before judging the new result.

Name the section that should've stayed put, and ask the AI to restore it while still applying your correction everywhere else.

Read the whole result again before accepting it.

Point out the detail that only belongs to this case.

Ask the AI to pull it from the Skill or project memory, then check the corrected instruction yourself.

Open the corrected instructions to check that the rule you did want saved is still there.