The AI personal assistant can turn plans and comparisons into public or private Files, extending its work beyond text and calls.
By [Ryan Merket](https://runtimewire.com/author/ryan-merket)
· Published
Primary source: [X](https://x.com/noahrshinn/status/2101080443667767385)
Why it matters #
Shareable Files turn Instinct's private agent work into distributable software objects, creating a potential growth loop while raising the cost of inaccurate or overly revealing output.
Noah Shinn (@noahrshinn) launched shareable Files for Instinct on September 18th, giving the AI personal assistant a way to turn its work into private or public interactive pages that users can send to other people.
https://x.com/noahrshinn/status/2101080443667767385 Instinct will create a File when a request calls for a structured artifact, such as a comparison, a plan or an explanation meant for someone else, Shinn said in a post on X. The feature moves Instinct's output beyond the text messages and phone calls that have defined the product so far.
The distinction matters for Shinn's product strategy. An assistant that books a flight or cancels a subscription can complete the job inside a private conversation. Planning a trip, evaluating a purchase or coordinating a project often produces information that needs to survive after the chat ends and travel to people who never interacted with the agent. Files give Instinct an output layer for that work.
Shinn describes Files as interactive pages rather than exported documents. Instinct can decide when to create one as part of handling a request, preserving the company's broader pitch that users should delegate outcomes rather than operate a collection of AI tools themselves.
That approach follows Instinct's deliberately plain interface. Instinct says users can text or call the assistant, which is trained to operate phones and computers and can connect to email, messaging, screens, audio and location. Its website describes an agent capable of following up on abandoned conversations, arranging rides and booking household services.
Shinn is building beyond the chat window
Shinn came to consumer AI agents through research on how language models use tools and improve across repeated attempts. According to his personal site, he was one of the first employees at enterprise AI agent developer Sierra and previously worked on type prediction and code generation in Northeastern University's Programming Language Group.
He was also a co-author of Reflexion, a NeurIPS 2023 paper that proposed using verbal feedback to help language agents learn from earlier failures. Instinct applies that research background to everyday work: the assistant receives a loosely defined request, gathers context, uses external services and returns a result through familiar communication channels.
Files extend that sequence through the final handoff. The result can become a destination of its own, available to the user privately or published for others. That gives Instinct a surface closer to a lightweight document or application while allowing Shinn to keep the assistant itself inside messaging and voice interfaces.
The launch also makes presentation part of the agent's job. A useful comparison page must organize choices clearly, preserve the relevant context and remain accurate when another person opens it without seeing the conversation that produced it. Instinct's own terms of service tell users to verify AI output and acknowledge that the assistant can produce incorrect or incomplete information.
Public Files raise the stakes further because an error can leave the original conversation and reach a wider audience. Instinct's terms place responsibility for output use and sharing settings on the user, while its privacy policy says the assistant may access documents, private communications and connected applications depending on the permissions granted. Users deciding whether to make a File public will need to check which personal details the agent carried into the page.
A heavily funded assistant adds a distribution surface
Instinct introduced Files less than a month after disclosing a large financing. On August 26th, TechCrunch reported that Spear Street Technology, the corporate entity behind Instinct, had raised a $250 million Series B co-led by Index Ventures and Benchmark. The round valued Instinct at $2.5 billion and brought its reported total funding to $350 million.
At the time, Instinct was operating through an invite-only beta and was already drawing scrutiny over the breadth of access required to complete tasks across users' accounts. Instinct's privacy policy, revised August 26th, says the assistant may process messages, emails, documents, payment information, credentials, location and other sensitive data when users grant the relevant permissions. It also provides a model-training opt-out, with stated exceptions for information flagged for safety review.
Shareable Files give Shinn another way to demonstrate the product without requiring every viewer to grant that access. A user can ask Instinct to research or organize something, then distribute the result as a page. Each shared File can effectively serve as a product demonstration generated from a real task, putting Instinct's work in front of prospective users through the people already testing it.
That distribution loop is valuable for an assistant whose usefulness depends on trust. The pages make Instinct's output easier to inspect and circulate. They also expose the quality of its reasoning, source selection and handling of private context outside the controlled confines of a one-to-one conversation. Shinn is betting that Instinct's work can stand on its own once it leaves the chat.