Noah Shinn's Instinct seeks at least $200M at a valuation above $2.5B

Why it matters

Investors are valuing Instinct like a breakout consumer platform before a public launch. Shinn's real test is turning agent reliability research into durable user trust.

Noah Shinn's Instinct is in talks to raise at least $200 million at a valuation above $2.5 billion, Forbes reported Wednesday, four months after the former Sierra researcher founded the personal AI assistant.

Benchmark and Index Ventures are discussing leading the financing, according to Forbes, which cited several people familiar with the deal. The round has not been reported as closed. If completed on those terms, it would value Instinct at roughly five times the level set during a $75 million Series A led by Kleiner Perkins earlier in August.

Shinn has spent much of his short career studying the problem Instinct now has to solve: getting language models to take actions reliably. He joined Sierra as one of its first employees after research work in programming languages, machine learning and computational photochemistry at Northeastern University and MIT.

His published work includes Reflexion, a framework that lets language agents improve through written self-reflection, and tau-bench, an evaluation for agents interacting with users and software tools while following business rules. Early tau-bench tests found that leading agents completed fewer than half of tasks and became substantially less reliable across repeated attempts.

That background gives Shinn a credible route into a crowded agent market. It also sets a demanding standard for an assistant being entrusted with inboxes, calendars, payment details and personal messages.

A private assistant gets a public-market-sized price

Forbes reported that Instinct is available only by invitation from venture capital insiders.

The product connects to email, messaging, location, screen and audio data. Users communicate with it through text messages, calls or WhatsApp and can ask it to arrange airport rides, make restaurant reservations, book travel, manage an inbox or complete purchases. Instinct describes the product as a personal assistant that operates phones and computers in much the same way a person would.

That interface is Shinn's distribution bet. Earlier autonomous agents often required users to install software, configure models and maintain infrastructure. Instinct packages computer use inside familiar communications channels, turning a technically complex system into a contact that can be texted.

Investors have been paying for that simplicity before Instinct has opened access broadly or begun charging users. Forbes reported that Sarah Guo of Conviction and Neil Mehta of Greenoaks led an earlier financing at a valuation above $100 million. Mamoon Hamid of Kleiner Perkins then led the $75 million Series A at a valuation above $500 million in early August.

The proposed Benchmark and Index round would take Instinct above $2.5 billion within weeks of that Series A and to roughly 25 times its first reported valuation. Forbes reported at least three financings over the past few months, though the precise total raised has not been established.

Instinct is currently free. Forbes reported that possible business models include a subscription for a digital chief of staff or advertising informed by the data Instinct can access. A subscription would align revenue with users; advertising would make the depth of Instinct's access central to both personalization and monetization.

Trust is already part of the product

Instinct's ability to act makes it useful, but each additional permission increases the cost of a mistake. TechCrunch reported on August 24 that testers had raised concerns about unauthorized actions, phishing risks and data retained after connected accounts were disconnected.

The report also cited an earlier version of Instinct's terms containing a broad, perpetual and irrevocable license covering user materials. Instinct's current terms, marked as revised on August 26, do not contain that phrase. They now provide a model-training opt-out and a process for requesting deletion of indexed data from connected services.

The revised language still gives Instinct extensive operating room. Instinct may train models on user materials unless users opt out, and the opt-out applies going forward. Instinct can retain indexed data after a service is disconnected unless a user separately requests deletion. The terms also appoint Instinct as a user's agent for entering agreements, commitments or transactions that can be binding on the user.

Forbes reported that AI founder Claire Vo said Instinct continued accessing her Gmail and retained copies of emails containing financial data after she disconnected her Google account.

Forbes cited investor Jason Yeh, who said Instinct booked a restaurant table carrying a $200 cancellation fee after he asked it to look for availability. Other testers described the product sending an email without approval or continuing to summarize stored messages after a Google account had been disconnected.

Shinn built his research reputation around measuring where agents fail across repeated, tool-heavy tasks. Instinct turns that research problem into a consumer contract: users grant access because the assistant saves time, then continue granting it only while Instinct behaves predictably.

The proposed valuation shows how aggressively investors are pricing the possibility that Shinn can make autonomous agents ordinary. The harder milestone will arrive when Instinct moves beyond a small group of technology investors and founders, whose tolerance for experimental software is unusually high. Instinct will have to earn the same permissions from consumers who expect a personal assistant to understand the difference between finding a reservation and booking one.