Delegation guide
Using Instinct AI: five practical briefs for everyday tasks
Turn ‘help with my life’ into a clear request: what to look at, what may change, when to check back, and how to prove the job is done.
Key takeaways
- The five briefs are original examples to adapt after confirming account capabilities.
- Specify the information, action boundary, deadline, and completion evidence in every request.
- Treat approval in a prompt as an instruction; also inspect the product's actual permission controls.
- Judge usefulness by completed tasks and reduced review work, not the number of messages received.
A task brief starts with a result you can recognize
These are proposed workflows, not transcripts or promises about Instinct's current capabilities. Confirm that your account supports the necessary tools before trying them. A useful task brief names the goal, the information the assistant may use, the actions it may take, the deadline, and the evidence you expect back.
For example, ‘sort out my week’ leaves the assistant to guess what matters and what it may change. ‘Draft a plan around these fixed commitments; leave the calendar unchanged until I review it’ gives you a specific result to inspect. More precise instructions also make a failure easier to diagnose: missing context, a mistaken interpretation, or an action that exceeded the brief.
Instinct's terms say actions may have consequences and that safeguards do not guarantee protection from unintended actions. Pair your instructions with the controls actually available in the connected service. A sentence asking for approval is not a technical guarantee that an account cannot be changed.
1. Find the personal follow-ups that need a decision
Suggested brief: ‘Review the personal threads I select from this week. List where someone appears to be waiting for my answer. For each one, show the source, what is being requested, and a suggested next step. Draft replies only; ask before sending anything.’
This gives you a short queue to review without granting blanket authority over every conversation. Include examples of messages that do not need a reply, such as newsletters or a finished exchange. A recent message and an unresolved request are different things, and the assistant needs enough context to distinguish them.
Define completion as an accurate list with links back to the conversations. After review, approve individual replies or close the items yourself. Keep a note of false positives: they reveal whether the assistant is reacting to keywords or understanding the commitments you actually made.
2. Turn a busy day into a proposed plan
Suggested brief: ‘Use tomorrow's calendar and this task list to propose a realistic day. Keep confirmed appointments fixed. Reserve travel and lunch time, and show which tasks will not fit. Do not edit events yet. Send the plan by 7 p.m. in my current time zone.’
The most useful result may be a task the assistant recommends deferring. If every item is squeezed into the schedule, the plan has not resolved the tradeoff. Give estimates for unfamiliar tasks and specify the commitments that cannot move. Review buffers around appointments that involve travel or preparation.
Check the proposal against your original calendar. The date, time zone, event ownership, and attendee list matter more than a persuasive explanation. If you decide to create or move events, make that a separate instruction and inspect the saved events afterward.
3. Prepare a ride plan before authorizing a booking
Suggested brief: ‘Using this itinerary, suggest when I should leave for the airport and compare the transport options I specify. Show pickup location, estimated total cost, arrival buffer, and any information that is missing. Ask me before booking or sharing my details with a provider.’
This brief divides planning from purchase. An assistant may need the flight date and terminal to propose a departure time; it does not need permission to charge a payment method just to prepare that proposal. Ask it to distinguish a current quote from an estimate and to surface uncertainty in travel time.
If you later authorize a booking, identify the provider, pickup date and address, passenger count, acceptable cost, and what to do when the quoted option is unavailable. Completion means the provider's confirmation matches the request. A message saying ‘all set’ is not enough to establish that a ride exists.
4. Organize a household service request
Suggested brief: ‘Help me prepare a request for this household repair. Summarize the issue from the details I provide, list questions a service provider is likely to ask, and draft an inquiry. Do not contact anyone, share my address, or accept an appointment until I approve the details.’
A useful first result is a clear description of the job, not an immediate appointment. Include what you have observed, any access constraints, suitable time windows, and whether another household member needs to be present. Leave diagnosis or specialized judgments to the relevant professional.
If communication becomes part of the task, specify the recipient and the approved message. Then keep the quote, scope of work, arrival window, and cancellation conditions together. The job is ready to schedule when those details are resolved; a vague expression of availability is only an intermediate step.
5. Make proactive updates earn their interruption
Suggested brief: ‘For the commitments on this list, tell me when something changes that requires my decision. Send a single evening summary for routine progress. Use an immediate message only if a deadline is within two hours and an action is still needed. Stop this routine on Friday.’
This defines a reason to interrupt, a summary schedule, and an end date. Set those expectations before asking for persistent monitoring. Also confirm whether the product can enforce them and which notification settings control the behavior. Without that check, a preference in a conversation may not affect every channel.
Review the messages after a few days. Did they arrive early enough to act? Did they repeat information you had already resolved? Did silence mean nothing changed, or that the routine had stopped working? Ask where you can inspect the routine's status, and keep the ones that reliably reduce the need to check manually.
Finish with a review and a way to reset access
For each workflow, record whether the result was correct, how many corrections it required, and how much effort you spent reviewing it. This is a personal usefulness check, not a leaderboard. Repeat tasks only when the total work is smaller than doing them yourself and the remaining uncertainty is acceptable to you.
When a task ends, revisit its connections and recurring instructions. Instinct's privacy policy says disconnecting a third-party integration does not automatically delete previously collected data. Use the documented Workspace deletion controls when removal of indexed information is also your goal.
Keep the brief and an example of a good result somewhere you control. That makes it easier to explain a correction, restart a task with different permissions, or evaluate another assistant without rebuilding your process from memory.
Directory picks
Agents mentioned in this guide
Source notes
Primary references
We use first-party product pages and technical documentation wherever possible. Listing access can change, so check the linked source before relying on a number or feature.
Related resources