Short answer: Instinct is designed to do more than answer questions. Its official site gives examples such as following up on dropped threads, proactively calling or texting, arranging an airport ride, and booking a handyman. Its Terms also describe the service taking actions through connected services on a user's behalf.
The wider list of tasks is broader, but the evidence is not equally strong for every category. That is why this site separates official, independent-report, and proposed-needs-verification records instead of publishing one unqualified capability list.
What tasks does Instinct officially show?
The strongest first-party examples come from Instinct's own homepage. Current examples include:
- following up on dropped threads;
- proactively calling or texting;
- arranging an airport ride;
- booking a handyman.
The Terms provide the general action model behind those examples. They describe Instinct interacting with connected services and taking actions for the user, subject to the permissions the user grants.
That establishes that action-taking is part of the product design. It does not prove that every conceivable errand is available to every account.
What other task categories have been reported?
Reuters and TechCrunch have described or reported Instinct handling a wider range of work, including:
- trip planning and travel booking;
- restaurant reservations;
- purchases and groceries;
- tickets;
- bills and subscription-related tasks;
- research and other errands.
TechCrunch also reported features that expand the kinds of tasks Instinct can attempt. Those include an Instinct email address for account creation and business contact, location sharing for nearby places and itineraries, Stripe-supported payments, and an early-access Concierge feature that could make phone calls for tasks such as restaurant booking or appointment waitlists.
These are useful reports, but they remain reported capabilities rather than a promise that every account has every feature.
Why the Task Database uses evidence levels
A statement like "Instinct can book flights" hides several separate questions:
- Was the behavior shown by Instinct itself?
- Was it reported by a reputable publication?
- Did a tester actually complete the task?
- Did the task make a mistake?
- What permissions were required?
- Could the action cost money or be difficult to reverse?
The Instinct Task Database stores those details per task.
For example:
- Arrange an airport ride has official support from Instinct's own examples.
- Book a flight is supported by reporting and hands-on testing, but the same testing documented a booking error.
- Book a hotel has successful and unsuccessful evidence, including an incorrect-rate test.
- Request an airline refund has a successful independent test, not a universal refund guarantee.
The evidence level is therefore as important as the task name.
Travel tasks
Travel is one of the most heavily documented Instinct categories.
Current evidence covers trip planning, flights, hotels, rides, itineraries, fees, bookings, and airline refunds. Founder comments reported by TechCrunch also indicate travel is a major transaction category on the platform.
Start with Instinct AI Travel, then use the specific Flights and Hotels pages for their separate reliability limits.
Shopping and purchase tasks
Instinct's Terms explicitly contemplate purchases made through connected services. Independent tests and reporting also cover product research and real purchases.
The important distinction is between:
- researching products;
- recommending products;
- preparing a purchase;
- actually authorizing a transaction.
Risk rises sharply at the final step.
The Instinct AI Shopping guide explains the transaction boundary, while task records such as Compare products expose the evidence and permissions for individual workflows.
Refunds and subscription tasks
Independent testers have reported successful refunds, including an airline refund and a forgotten-subscription refund. Other tests show why this category needs supervision: WIRED documented a DoorDash cancellation that was executed despite the user's refund condition and resulted in a loss.
See Instinct AI Refunds for the evidence and limitations.
Phone calls and offline tasks
TechCrunch reported that Instinct Concierge could make calls for tasks that are difficult to complete entirely online, such as phone-only restaurant reservations and appointment waitlists.
The feature was described as an early-access rollout. We therefore do not treat phone calling as universally enabled on every account.
What permissions can tasks require?
Permissions depend on the action.
A task may require some combination of:
- trip dates and traveler details;
- location;
- email or calendar context;
- account access;
- payment information;
- a merchant or airline account;
- permission to call or message a third party.
A request that sounds simple can become high-risk once it requires payment or changes an external account.
The Task Database shows the expected permissions and risk level for each record.
Does Instinct always complete a supported task correctly?
No verified source establishes that guarantee.
Hands-on testing has documented both successes and failures. Business Insider reported useful product research and refunds but also harmful mistakes. WIRED reported a successful airline refund and phishing warning, but also a conditional cancellation that went wrong.
Instinct's own Terms warn that outputs and actions can be incorrect or incomplete.
For this reason, "supported" should never be interpreted as "guaranteed accurate."
How to write a safer request
For high-impact tasks, ask for checkpoints.
A stronger request usually includes:
- the exact desired outcome;
- constraints such as budget, dates, or merchant;
- a requirement to show the final price or terms;
- an instruction to stop before charging, canceling, sending, or deleting when practical;
- a request for a confirmation number or final status.
For example, the Task Database's flight-booking record suggests asking Instinct to show the final fare and restrictions and to wait for approval before purchase.
Frequently asked questions
What can Instinct AI do?
Official examples include follow-ups, proactive calls or texts, arranging an airport ride, and booking a handyman. Reporting and hands-on tests cover a wider range of travel, shopping, refund, restaurant, appointment, subscription, and research tasks.
Can Instinct make purchases?
Its Terms explicitly allow actions that include purchases through connected services. Users remain responsible for merchant transactions and should verify the final amount and terms.
Can Instinct make phone calls?
TechCrunch reported an Instinct Concierge rollout that could place calls for certain tasks. Availability was described as early access, so it may vary.
Are all tasks in the database guaranteed features?
No. Each record has an evidence level. Some are official, some are supported by independent reporting, and some remain research-only until stronger Instinct-specific evidence appears.
Where can I browse tasks?
Use the searchable Instinct Task Database to filter by category, permission, risk, and evidence level.