Is This Task Safe to Hand to AI? An 8-Question Self-Test

Strategy

Oct 26

A task is safe to hand to AI when three checks are in place: a human who approves it, guardrails written down ahead of time, and a record you can go back to. My rule is simple: "AI can propose, but I ratify." This self-test scores any task you are thinking about handing off, in eight questions, and tells you what to do next.

For coaches, consultants, and entrepreneurs who are using AI and are not sure which jobs to hand over. Use it before you set up a new automation, or before you let a tool do anything in your name.

Can I explain this out loud? If a colleague asked how this task works, could you explain what the AI does at each step and why?
No, I would be guessing Roughly, but not the details Yes, step by step
Where does this leave my hands? When the AI finishes, where does the work go?
Straight out to other people (sent, posted, published) It waits as a draft, but I often send without reading every word It stays with me until I read the words and approve them
Have I checked the facts at the source? Names, numbers, dates, claims. How do they get checked?
I trust it because it sounds confident I spot check now and then I check them at the source, or I keep the real version safe and the tool flags changes instead of making them
Has it earned more rope? How long has the AI done this exact kind of job for you?
This is the first time A few runs, with mixed results A track record of small jobs it got right, where mistakes were easy to undo
If it gets this wrong, how hard is it to take back? Picture the worst version of the output going out.
Hard or impossible (public, sent, deleted, or paid for) I could fix it, but it would take effort and an awkward message Easy. It is private or reversible and nobody outside sees it first
Whose words and information is it working from? Think about what the AI reads and who ends up receiving the result.
It speaks as me to someone who does not know me, or it touches information that is not mine to share A mix of my material and outside material Built from my own words and my own documents, with nothing sensitive in it
Are the directions written down ahead of time? Before the AI starts, does it have written rules for this job?
No, I explain it fresh in chat each time Some of it is written, some lives in my head Yes, the directions live in a document I can edit and point it to
Is there a paper trail I can go back to? If something looks off next month, can you see what happened?
No record, or only a chat I would never scroll back through Some record, scattered across tools Yes, a log in a place I control that I can read in order

Keep it in your hands (0 to 5)

This one is not ready to hand off. Use AI to think out loud, sort your notes, or draft a first pass, and keep every decision and every send with you. Next action: pick the question where you scored a 0 and fix that one thing before you try again, or choose a smaller job.

Draft only, and you press send (6 to 11)

AI can read, sort, log, and draft here. It should not send, post, publish, or delete. Next action: write the directions down in one document and start a simple record of what it did, then score this task again.

It has earned a little more rope (12 to 16)

You have the three checks in place: a human, guardrails, and a record. Hand it off, keep approving the final step, and keep the paper trail. Next action: write down what makes this job safe so the next job can be scored the same way.

FAQ

How do I know if I can trust AI with a task?

Ask four questions. Can I explain this out loud? Where does this leave my hands? Have I checked the facts at the source? Has it earned more rope? Work that is private, reversible, or built from your own words is a safer place to start. Work that is public, hard to take back, or speaks as you to someone who does not know you needs more caution.

What should AI never do without me?

In my setup, my AI assistant can read, sort, log, and draft. It cannot send, post, publish, or delete. The moment of sending is where I slow down and read the words myself.

Can I trust AI to check its own facts?

No. Confidence isn't a citation, so the facts get checked at the source. One practical move is to keep the real version of a document somewhere safe and have the tool flag changes instead of making them.

How do I start handing work to AI without taking a big risk?

Nobody gets the keys to the safety deposit box on day one. Start with small jobs where a mistake is easy to undo, and let the AI earn more rope from there. If you catch yourself giving the same feedback twice, it probably belongs in a document, not a chat.

Go Deeper

The episode goes through the Skynet Check, the four questions behind the first half of this test, and the stories that shaped them. Listen to Ep 38 (Skynet Check)

Link: https://jesswebber.com/bigideasmadesimple/how-much-trust-ai

Ep 27: The Trust Test: https://jesswebber.com/bigideasmadesimple/trust-tell-managing-people
Ep 22: Your Brain Isn't A Filing Cabinet: https://jesswebber.com/bigideasmadesimple/brain-not-a-filing-cabinet
Ep 20: You Were Never Supposed to Do it All: https://jesswebber.com/bigideasmadesimple/never-supposed-to-do-it-all

Big Ideas Made Simple is a decision-making podcast for fast thinkers who are tired of hiding behind hustle and perfection. Hosted by Jess Webber. New episodes weekly at bigideasmadesimple.captivate.fm.

Get the next one