# Benjamin (Ben) Mann

> Benjamin 'Ben' Mann is an Anthropic co-founder, engineer, and AI researcher whose route to frontier AI ran from Columbia and six years at Google through a startup attempt, a consequential reading of Nick Bostrom's Superintelligence, work at MIRI and OpenAI, and the founding of Anthropic in 2021. A co-author of the GPT-3 paper and later Anthropic's product engineering lead, he now builds experimental products in Anthropic Labs alongside Mike Krieger while arguing that capable AI and serious safety work must be developed together.

- **Role:** Co-founder and Anthropic Labs co-lead at Anthropic
- **Organizations:** Anthropic, OpenAI, Machine Intelligence Research Institute, Google
- **Education:** BS in Computer Science, magna cum laude, Columbia University - Fu Foundation School of Engineering and Applied Science
- **Known for:** Co-founded Anthropic in 2021., Was the second-listed author of the 2020 GPT-3 paper 'Language Models are Few-Shot Learners.', Implemented large-scale models, training infrastructure, and model-parallel strategies for GPT-3 with other OpenAI researchers.

## Career timeline

- **2010** — Built an early offline Chinese character recognition and translation app for Android with his brother and classmates; contemporary phones were not powerful enough for the idea to continue.
- **2011-2017** — Spent six years as a software engineer at Google, working across UX and infrastructure, serving as a tech lead/manager, and helping build Waze Carpool.
- **2017** — Published an essay after nine months immersed in AI research, writing that the experience had changed how he saw the world and the role he wanted to play in it.
- **2018-2019** — Pursued startup work and AI-safety research, including work connected with the Machine Intelligence Research Institute.
- **2019-2020** — Worked at OpenAI on GPT-3 infrastructure, efficiency, safety, training systems, and model-parallel implementation.
- **2020** — Named second author on 'Language Models are Few-Shot Learners,' the GPT-3 research paper.
- **2021** — Co-founded Anthropic with former OpenAI colleagues.
- **2023** — Publicly described himself as leading product engineering at Anthropic and working to create a 'safety race.'
- **2025** — Led Anthropic Labs, home to experimental work around Claude Code and the Model Context Protocol, while discussing Claude 4, agents, and safety in public interviews.
- **2026** — Anthropic expanded Labs and announced that Mike Krieger would build alongside Mann on experimental products at the frontier of Claude's capabilities.

## Achievements

- Co-founded Anthropic in 2021.
- Was the second-listed author of the 2020 GPT-3 paper 'Language Models are Few-Shot Learners.'
- Implemented large-scale models, training infrastructure, and model-parallel strategies for GPT-3 with other OpenAI researchers.
- Led product engineering at Anthropic and later Anthropic Labs.
- Co-authored Anthropic research on language-model behavior, interpretability, repeated data, red teaming, and AI feedback.
- Helped create the product-building approach that produced Claude Code and the Model Context Protocol.

## Latest updates

- **2026-01** — Anthropic formally announced the expansion of Anthropic Labs, with Mike Krieger joining to build alongside Ben Mann.
- **2025-07** — Appeared on Lenny's Podcast for a wide-ranging discussion of leaving OpenAI, the pace of AI progress, jobs, agents, alignment, and preparing children for an AI-shaped future.
- **2025-06** — Appeared on No Priors to discuss Claude 4, long-horizon coding agents, RLAIF, MCP, and Anthropic's safety thresholds.

## Links

- Website: https://benjmann.net/
- LinkedIn: https://www.linkedin.com/in/benjamin-mann/
- Twitter/X: https://x.com/8enmann
- GitHub: https://github.com/8enmann

---

Profile page: https://yespress.io/benjamin-ben-mann
Published by YesPress — https://yespress.io
Last updated: 2026-09-26