Short answer: Instinct works by combining a conversational interface with connected-service context and the ability to take actions for the user. Its Terms describe an AI personal assistant that accepts input, generates output, connects to third-party services, and can perform actions through those services.
The important distinction is that Instinct is not limited to producing advice. Depending on the permissions and services involved, it can move from understanding a request to actually interacting with an external account or business.
The basic Instinct workflow
A useful way to understand Instinct is as a five-step loop:
- You give Instinct a goal.
- Instinct gathers relevant context from the information and connected services you have made available.
- The agent plans a path toward the requested outcome.
- It uses tools or connected services to carry out parts of the task.
- It reports back, asks for input, or continues the task depending on what is required.
Instinct's Terms explicitly distinguish Input, Output, and Actions. That matters because an action can have consequences outside the chat itself.
For examples of individual actions with evidence and risk labels, browse the Instinct Task Database.
Step 1: you give Instinct a request
Instinct is designed around natural-language requests rather than a fixed menu of workflows.
The official product positioning emphasizes a personal assistant you can reach through conversation. Recent reporting has described text-based use and newer calling capabilities, but the exact interface can change as the product develops.
A request might be informational, such as asking for research, or action-oriented, such as asking the assistant to arrange something on your behalf.
Step 2: connected services provide context
Instinct becomes more useful when it can work with data you have authorized.
Its Privacy Policy says the assistant may receive information from connected applications and services, private communications, documents, location, account information, and other data depending on the permissions granted.
The policy separately identifies Google Workspace services including:
- Gmail;
- Google Calendar;
- Google Drive;
- Google Docs;
- Google Sheets;
- Google Slides;
- Google Tasks.
That does not mean every Instinct account automatically exposes all of those services. Access depends on what you connect and authorize.
See Instinct AI Integrations for a service-by-service overview.
Step 3: Instinct plans and acts
Instinct's Terms describe the service as taking actions that are responsive to the user's input. The Terms also authorize interaction with connected third-party services.
This can include practical steps such as:
- reading or using contextual information;
- sending information to a connected service;
- completing a booking workflow;
- communicating with another party;
- making a purchase on the user's behalf;
- accepting third-party terms as the user's agent when a task requires it.
This is why Instinct's risk profile is different from a normal answer-only chatbot.
If the assistant makes a wrong recommendation, that is one kind of failure. If it sends a message, changes a booking, or spends money incorrectly, the consequences can be more direct.
Step 4: credentials, accounts, email, and payments extend what the agent can do
Recent product reporting shows that Instinct has been adding infrastructure designed to reduce the number of times a user must manually intervene.
TechCrunch reported that Instinct introduced its own email addresses so the agent could create or manage accounts, contact businesses, and follow up on tasks. The same report discussed support involving 1Password for logins and Stripe for payments.
Those reports help explain the product direction, but they should not be generalized into a guarantee that every account has every integration.
The official Terms remain the stronger source for the general rule: Instinct can interact with connected services when the user authorizes that access.
Step 5: the user still needs to verify important outcomes
Instinct's own Terms say the service is not error-free. Outputs may be incorrect or incomplete, and actions may not always be reversible.
The Terms also put responsibility on the user to evaluate whether the service is appropriate for a task and to review high-impact actions where meaningful human review is necessary.
That means "autonomous" should not be interpreted as "safe to ignore."
For money, travel, appointments, account changes, or external communications, verify the final state:
- Was the booking actually confirmed?
- Was the correct amount charged?
- Was the message sent to the right recipient?
- Was a cancellation really completed?
- Did the third-party service accept the change?
Our Instinct AI Safety guide covers these risks in more detail.
Does Instinct use one AI model?
The public evidence we verified does not establish a single foundation-model provider or a fixed model stack for every task.
That is important because agent products often combine multiple systems: language models, browsing or computer-use tools, retrieval, integrations, payment systems, credential systems, and workflow logic.
We therefore do not infer a hidden model architecture from investors, partners, or outside speculation.
Is Instinct always running in the background?
The official Privacy Policy says the personal assistant is designed to be available when engaged. Recent reporting also describes follow-up behavior and tasks that can continue until the agent needs user input.
That supports the general idea of a persistent assistant, but it does not establish a universal promise that every task runs continuously or that every account has the same background capabilities.
How permissions affect what Instinct can do
Permissions are one of the most important parts of how Instinct works.
A task that only needs public web research may require little private data. A task involving email, calendar, payments, or account logins may require much more access.
The capability of the assistant is therefore partly constrained by:
- what services are connected;
- what permissions the user grants;
- what information is available;
- the third party's own rules;
- whether the task requires confirmation;
- the current product rollout.
Use Instinct AI Privacy for the current primary-source permission and data-access rules.
Instinct is an agent, not a guarantee
The cleanest mental model is:
request → context → plan → tool/action → verification
That model explains both the appeal and the risk.
Instinct can remove manual steps from everyday tasks, but the same ability to act means users should care about permissions, accuracy, reversibility, and the final result.
Frequently asked questions
Does Instinct control my accounts directly?
It can interact with connected services and accounts when authorized. The exact access depends on the service and permissions you grant.
Can Instinct make purchases?
Its Terms explicitly cover purchase actions through connected services. The user remains the buyer, and the third party's pricing, refund, and cancellation rules still apply.
Can Instinct read Gmail and Calendar?
Its Privacy Policy says that if you link Google Workspace, the service can receive access to Gmail, Google Calendar, and other named Workspace products.
Does Instinct need my passwords?
Some workflows may involve credentials or third-party login systems. Instinct's Privacy Policy describes account credentials as a possible category of information users may provide, while recent reporting has discussed 1Password-related login support.
Is Instinct fully autonomous?
It can take actions on the user's behalf, but the Terms explicitly warn that actions may be incorrect and not always reversible. Important outcomes should still be reviewed.