Noah Shinn

As first author, Noah Shinn published the paper “Reflexion: Language Agents with Verbal Reinforcement Learning” with Federico Cassano, Shunyu Yao, Karthik Narasimhan, and others, listing Northeastern University as his affiliation. The paper proposed the Reflexion framework: instead of updating model weights, language agents verbally reflect on task feedback and store that reflective text in an episodic memory buffer to improve decision-making in later attempts; it reported a 91% pass@1 accuracy on the HumanEval coding benchmark, above the 80% of GPT-4 at the time.⁠[1][1][1][1][1][1]

Shinn worked at Sierra, the AI company led by Bret Taylor. In June 2024, he and Shunyu Yao, Pedram Razavi, and Karthik Narasimhan published the agent benchmark τ-bench under Sierra’s name: it uses language models to simulate users in dynamic conversations with language agents equipped with domain-specific API tools and policy rules, evaluates them by whether the database state at the end of the conversation matches the goal state, and introduces a pass^k metric to measure reliability over multiple trials. The paper reported that even state-of-the-art function-calling agents such as gpt-4o succeeded on fewer than 50% of the tasks.⁠[2][3][3][3][3][3]

Around 2025, Shinn founded the personal agent company Instinct (operated by Spear Street Technology Inc.Instinct (AI Company)Instinct is a San Francisco-based artificial intelligence startup (legal entity Spear Street Technology Inc.) founded and led as chief executive by Noah Shinn. It develops a personal agent of the same name: users communicate with it by text or phone, and it uses its own phone and computer to order meals, book tickets, shop, cancel subscriptions, and more on their behalf. Between August and September 2026, the company raised funding at valuations of $2.5 billion and then $10 billion within about a month.Open the full entry) and became its chief executive; as of August 2026, he was 23. In September 2026, he said Instinct was a company he had started about a year earlier, and called the field “probably the most exciting software race ever.”⁠[4][2][5][6]

Instinct raised a $250 million Series B at a $2.5 billion valuation. Shinn posted on X that early users had used Instinct to plan cross-country road trips, buy groceries and concert tickets, cancel subscriptions, and even plan a wedding.⁠[4][4][4][4]

Instinct raised a $1 billion Series C at a $10 billion valuation. Shinn said the company was building the best personal agent able to handle the deeply personal nuances of everyday life, and that the funding would help bring Instinct to more people and continue building the future of personal AI.⁠[7][7][7]

Shinn was interviewed on episode 493 of Invest Like the BestInvest Like the BestInvest Like the Best is a business and investing interview podcast hosted by Patrick O’Shaughnessy. Launched in 2016, it features conversations with investors, founders and experts about their ideas, methods and stories; the show is now part of the Colossus podcast network.Open the full entry, hosted by Patrick O’ShaughnessyPatrick O’ShaughnessyPatrick O’Shaughnessy is an investor and podcast host who has hosted the interview show Invest Like the Best since 2016, founded the early-stage investment firm Positive Sum in 2020 and serves as its CEO, and is also founder and CEO of the media company Colossus.Open the full entry, who noted it was the first time Shinn had talked about the business in long form. He said Instinct was growing by word of mouth without any marketing spend, with five invites per user and the user base growing about 10% a day, to the point that some people were reselling invites on eBayeBayeBay Inc. is an online marketplace company headquartered in San Jose, California, United States, founded in 1995 by Pierre Omidyar; it started out with person-to-person online auctions and listed on Nasdaq in 1998. As of the end of 2025, it had 135 million active buyers, and its platform’s gross merchandise volume in 2025 was nearly $80 billion.Open the full entry for about $300; he said he spends a large share of his time on compute planning, and that the company could deliver performance comparable to frontier models at a very low cost.⁠[6][6][5][5][5][5][6][5]

He said users take several weeks to build trust in an agent: after about three weeks, about 40% of users had shared a personal credit card with Instinct. On the business model, he said he did not want Instinct to influence users’ behavior against their wishes and therefore would not build an advertising model; the product is free for users, and the company charges a transaction take rate to merchants that gain distribution through the platform, through which more than $1 billion a year in transactions now flows, about half of it from travel.⁠[5][5][5][6][5][5][5][5][5]

On pricing, Shinn said all users currently use the product for free, and that delivering it free for everyone for a lifetime is his personal goal, though he is “not going to commit to free for lifetime for now.”⁠[5][5][5]