What is Instinct? The personal assistant behind the name

Instinct wants you to give an assistant a task by text or phone, then let it work through the services involved. The useful question is how much of that task you can safely delegate.

What the company says

Instinct’s website describes a personal assistant that can use a phone and computer. Its examples include arranging appointments, ordering groceries and planning travel. These are the company’s advertised capabilities; we have not independently completed those tasks with the product.

Sequoia’s company profile identifies Noah Shinn as its founder, lists 2025 as the founding year, and records the firm’s partnership in 2026.

Start with the right Instinct

This article covers the product at instinct.com. A matching product name is not enough to establish identity. In particular, an MCP memory tool described in a coding-agent catalog is not evidence about this consumer assistant’s features, licensing or availability.

When evaluating an agent, check the official domain and operator before carrying over a review, price or security claim from another listing. We keep the source links beside the claims for that reason.

What would make it useful?

Consider a small, bounded task: find three suitable restaurants for Friday, compare their availability, and prepare a recommendation. A useful assistant needs to carry the date, location, budget and preferences through every step. It also needs to distinguish a proposed reservation from a confirmed one.

The next test is a change halfway through. If the time moves, does the assistant update the plan? If a booking fails, does it explain the failure? If an action would spend money or contact someone, is the approval request clear enough to make an informed decision?

These are our proposed evaluation questions, not observations about Instinct’s behavior.

What we still need to verify

Before recommending a subscription or handing over an important task, we want to establish:

• Current pricing, access requirements and regional availability.
• The accounts and permissions a task needs.
• Where the assistant asks for approval and how that approval is bounded.
• Whether the result includes evidence that the task actually completed.
• How cancellation, failure recovery and access revocation work.

The public sources reviewed here do not answer those questions sufficiently for a recommendation. An accessible signup page would not, by itself, prove the full service works for every visitor.

Our first trial

Our first trial should use an ordinary task, a written success condition and a small permission scope. We will record setup time, intervention, the final result and any unfinished steps. Until that trial exists, this remains a source-based explainer.

Read how we evaluate agents for the standard we intend to apply.