During a 45-minute walk I used voice instructions, through my connected AI setup, to develop the slide deck for my automation masterclass and create the demonstration files. By the end of the walk those materials were made. I still had to verify the deck and test the demos before teaching.

Save the review for a separate sitting. My setup was already configured. A voice session reaches apps and files only through that kind of connection, and a phone by itself isn’t a link to a computer at home. I treat the walk as flexibility: one outline, then a real check later.

I used to think I had to write the implementation myself. Start with a familiar repetitive task. Describe what good looks like, in plain English. Ask the AI to propose a plan. Approve it before anything changes. You are specifying the result and checking the work. Code still sits in the setup. You may not be writing it by hand. Access, permissions, and scheduling depend on the actual setup.

Four questions before you automate anything

The best first automation is repetitive, familiar, and easy to check. Leave the most strategic task for later.

Can the AI access the inputs? It has to see the folder or the messages. Fix access before you polish the prompt.

Can you describe a good result? Write the result in plain English. For files, that means a proposed category and a clearer filename, with a date where a date helps.

Can you check the work in under two minutes? I use two minutes as a selection rule from the class. It chooses the task, separate from any performance claim. If the review would run long, pick something smaller.

Would a mistake be easy to reverse? Choose a task you can undo.

List three tasks you repeated this week. Choose the one that passes all four. Try it on a small folder of disposable sample files. I used roughly ten sample files in the live demonstration. I know. Ten files. Small enough to check every one.

Sound too small? Good.

Let the AI Show You the Plan Before It Touches a File

PAVE, then read the table

I use a loop called PAVE so there is a checkpoint before anything changes.

Point at the folder or the task.

Ask how the AI would handle it.

Verify the proposed plan yourself.

Execute only after you approve. Then check the result again.

On a folder, the first ask is an inventory. Nothing gets renamed, moved, archived, or deleted. For each file I request the current name, the contents, a proposed category, a clearer filename, and a confidence level. Uncertain cases get flagged. I read that table. Only after it looks right do I authorize dated filenames and category folders.

Duplicates and obsolete files get archived, so they stay available. Keep a backup too. Archiving preserves the file. A backup is what you restore from.

Run that inspection-only prompt on a copy of a cluttered folder. If a row looks wrong, ask for a revised plan.

Why one writer?

In the demo I let two AI tools inspect the same folder, because I wanted the comparison. Then I let only one of them make changes.

One writer at a time keeps the result understandable and keeps the two agents from interfering with the same files. I treated that as a precaution in the demonstration. Use the same precaution on a shared folder. Both tools can inspect. Only one reorganizes.

After the demonstration folder was organized, I asked a second AI tool to study the finished structure and use it as the model for a recurring cleanup. That tool had already inspected the original folder, so it had the earlier context and the organized result. The before-and-after gives the AI a model and gives you a standard for its rules.

Ask it to explain the rules it would infer before you apply them anywhere else. One well-organized sample teaches a pattern. It won’t correctly handle every future file. You still check the next batch.

Report first

A one-time cleanup still needs a supervisor once it becomes a daily routine. When I turned that folder cleanup into a daily routine, the AI offered three levels of autonomy. Report first and request approval. Act on the confident cases and flag the rest. Or act fully on its own.

Beginners should report first, so you can learn what the system does before you give it more freedom. In the demonstration that run was scheduled for 10:00 a.m. Those three levels were options in the workflow I showed. A local scheduled task may need the computer and the relevant app to remain running.

Bring the source to the decision

I compare the voice workflow to a short call with an executive assistant, the way I used to work. Ask what needs a decision. Hear the options. Give direction. The assistant carries out what you approved. Your judgment stays on the decision. The clicking and the organizing follow.

Try one brief voice review on a low-risk task, then look at the outcome. Use the shape of those assistant calls. A person who knows your work is more reliable than the AI. Voice still needs access configured before it reaches your files.

For email, I did this in a simulated inbox of eight dummy emails. I asked the AI to separate decisions and actions from reference material and from noise, and to show the full email body whenever a decision was needed. I proposed reviews at 11:00 a.m. and 2:00 p.m. Pair each decision with that source message. Keep sending and deleting behind an explicit approval.

Do this today

Pick one task you repeated this week that passes all four questions. Write the result in plain English. Point the AI at a copy, ask for the inspection or the plan, and approve it only after you have read it.

Recommended for you

Want the full system? 25X is the flagship productivity system we teach.

Explore 25X →

You may also Like

Last Updated: November 11, 2026

Read More

Last Updated: November 8, 2026

Read More

Last Updated: November 5, 2026

Read More

ABOUT THE AUTHOR

Thanh Pham

Founder of Asian Efficiency where we help people become more productive at work and in life. I've been featured on Forbes, Fast Company, and The Globe & Mail as a productivity thought leader. At AE I'm responsible for leading teams and executing our vision to assist people all over the world live their best life possible.


Leave a Reply


Your email address will not be published. Required fields are marked

{"email":"Email address invalid","url":"Website address invalid","required":"Required field missing"}