Nat Friedman

Nat Friedman served as GitHub CEO from 2018 through 2021 and announced GitHub Copilot’s technical preview during that tenure. His AI work also includes the NatBot browser-agent project, establishing AI Grant with Daniel Gross, and instigating, directing and sponsoring the Vesuvius Challenge to read carbonized ancient scrolls without opening them.[4][5][6][2][8][10]

On June 29, 2021, Friedman announced Copilot as an “AI pair programmer.” The preview drew context from the developer’s code to suggest whole lines or entire functions, with uses including writing tests and exploring unfamiliar APIs. GitHub developed it in collaboration with OpenAI, whose Codex system powered the suggestions. Friedman’s role here was GitHub’s CEO and public launch spokesperson; OpenAI created the underlying AI system.[5][4]

Friedman’s NatBot project connects GPT-3 to browser actions through a compact text interface. Its prompt supplies an objective, the current URL and a simplified description of the visible page. Formatting is stripped away, while links, buttons and inputs receive numbered identifiers. The model can issue commands to scroll, click an element, type into an input or type and submit a form.[6][7]

NatBot became an example for early LLM application tooling. In a September 2023 interview, LangChain founder Harrison Chase identified Friedman’s web agent as one of three end-to-end modules in LangChain’s initial October 2022 release, alongside LLM Math Chain and Self-Ask. Chase described using open source and academic examples to factor out common application patterns.[2]

Friedman’s AI funding work predates Copilot and NatBot: he and Daniel Gross established AI Grant in 2017. The initiative supported fundamental research through grants before broadening its focus toward AI products and startup investments, while retaining support for open source projects.[8][9]

The program page carrying a July 2023 application deadline explained the shift as a belief that AI research had become abundant while UX and product innovation were just beginning. It offered startups a $250,000 investment through a no-cap, no-discount MFN SAFE, distinct from a grant. Friedman and Gross jointly invested $10 million in AI Grant; AI Grant LLC made the company investments, and Hersh Desai was credited as organizer.[9]

AI Grant’s product-hunches section, dated August 2022, emphasized assistants that remove drudgery and products designed around imperfect models. Its instruction to “Design to minimize the user cost of bad model output” connected hallucination to interface design. These were the program’s stated views, without an individual byline.[9]

Friedman and Gross also supported Artificial Analysis through AI Grant and a subsequent seed investment. In a January 2026 interview, cofounder George Cameron credited both with helping the founders work methodically through difficult startup questions and decisions, calling them “great mentors.”[1]

In March 2023, the Vesuvius Challenge launched a competition to recover writing from carbonized Herculaneum scrolls. Friedman is its instigator, director and founding sponsor, alongside founding sponsor Daniel Gross and principal advisor Brent Seales. Giorgio Angelotti leads the project and technical team. The scientific foundation came from Seales’s University of Kentucky team, whose work demonstrated virtual unwrapping and detection of the scrolls’ elusive carbon ink.[10]

The challenge combines geometry reconstruction and ink detection. CT scans produce three-dimensional voxels; the writing surface must be segmented, meshed and flattened into a readable shape. Because carbon ink is nearly indistinguishable from papyrus in X-ray CT, models learn from fragments with visible ink and infer writing inside sealed scrolls. Iterative pseudo-labeling helps extend legibility, while surface tracing and generalization across scrolls remain open technical problems.[10]

The project reports that its initial prize—recovering four passages of 140 characters—was claimed within a year, and that PHerc. 1667 became the first Herculaneum scroll read end to end through virtual unwrapping in 2026. These are collective project achievements, with Friedman contributing direction and sponsorship.[10]

On his personal website, Friedman argues for speed because frequent contact with reality increases learning and forces attention onto what matters: “Time is the denominator.” He favors smaller teams for faster decisions, fewer meetings and more fun, and argues that exceptional individuals should be empowered to exercise their judgment. These are undated statements of his working philosophy.[4]