You said you’d need help with parts of this — want to take the first pass?
auto-dismisses · ✕
“Write the whole comparison report for me”
Fires when an Aid task gets fully handed over at low stakes. Fades on its own — nothing to click.
Interventions
Lightest to firmest, as they appear anchored to the chatbot’s composer. Every one has a way through.
Have you had a go at this part yourself? Learning sticks better that way.
Let me try firstSend anywaySnooze 1h
“Explain and solve this whole problem set for me”
“Let me try first” opens a box for your attempt — Manta appends it so the AI critiques instead of replaces.
You won’t easily be able to check this one — worth a sanity-check if you use it.
Got itSnooze 1h
“Write my entire lit review on attention economics”
Fires once per session when you hand over something you couldn’t check yourself — then silent, logged as unverifiable use on your dashboard.
You won’t be able to verify this, and it matters. Want a second pair of eyes before you use it?
Yes — help me check itSend as isSnooze 1h
“Draft the contract terms for our new supplier”
Never blocks — you chose this mode. The injection partly compensates for what you can’t verify.
You kept this one for yourself. Still want AI on it?
Continue anywayEdit promptSnooze 1h
“Decide for me: should I take the new job?”
Fires when a prefer-no-AI task gets a heavy hand-over. Stays until you act — never auto-sends. Continuing appends the guide-don’t-solve constraint.
AI isn’t permitted for this work. The prompt can’t go as written.
Edit promptCancelOverride with a reason (recorded)
“Write my assessed essay on the causes of WWI”
Rule-based domains block at any delegation level, either stakes. Overrides need a typed reason and appear in your weekly summary.