Theo Browne (T3)

Early life and education

Browne has said that he became interested in video through skateboarding, filming and editing skate videos, and that he was also closely involved in music.^[13] His LinkedIn profile lists Rensselaer Polytechnic Institute,^[14] and he has said that his degree is in audio engineering.^[13]

Career

Browne worked as a software engineer at Twitch.^[1][3] He has said that while Twitch was turning its video infrastructure into a product for Amazon Web Services, he felt that the WebRTC technology was not being given a fair hearing, and built a small demo app for putting several streamers in one window, from which Ping later grew.^[13] He told Business Insider that he had become frustrated with what he saw as a half commitment to new business areas, and that he quit after being reprimanded for work done during a hackathon that encroached on another team.^[1] He then worked briefly at another startup before founding his own.^[13]

Browne founded Ping Labs in 2021 with Mark Florkowski, who became its chief technology officer, and the company went through the Y Combinator accelerator in its Winter 2022 batch.^[1][4] Its first product, Ping, let streamers bring guests onto their broadcasts in higher definition and with lower latency, with subscriptions starting at $50 a month; its first client was the VTuber agency VShojo.^[1] Browne has said that he started his YouTube channel during Y Combinator, partly to learn about the creators Ping was made for.^[13] The company later released UploadThing, a file-upload service for full-stack web developers, in 2023,^[15] and PicThing, a background-removal tool for content creators.^[4]

The company's AI chat app, T3 Chat, offers models from several AI companies under one subscription,^[16][5] and news reports have described Browne as chief executive of the "chatbot platform T3 Chat".^[6] In August 2025 The Wall Street Journal quoted Browne, who can compare his costs for each model across thousands of user queries, as saying that "the arms race for who can make the smartest thing has resulted in a race for who can make the most expensive thing". He told the paper that most consumers use chatbots for tasks that do not need the most resource-intensive models, and that he was beginning to explore ways to nudge T3 Chat's users towards cheaper, "dumber" models.^[5] In March 2026 the company released T3 Code, a free, open-source app for running AI coding agents such as Anthropic's Claude Code and OpenAI's Codex, which Browne builds with the engineer Julius Marminge;^[17][16] its website says that more than 500,000 developers use it.^[18] In October 2026 Browne said that Anthropic's newest Claude Opus model had become the first to receive half of all prompts sent through T3 Code.^[7] In May 2026 he launched Lakebed, an agent-native command-line tool and hosting service for small TypeScript apps, as a free public alpha.^[19][20] Browne has also invested in other startups as an angel investor, including the text-to-3D design startup Adam.^[21]

T3 Stack

The "T3 Stack" is a web development stack made by Browne that its documentation describes as "focused on simplicity, modularity, and full-stack typesafety". Its core pieces are Next.js and TypeScript, almost always with Tailwind CSS, and with tRPC, Prisma and NextAuth.js for back-end work.^[8] The create-t3-app command-line tool, published on GitHub by the T3 Open Source organization since May 2022, sets up a new project from whichever of these parts a developer chooses; it had about 29,000 stars on GitHub as of October 2026.^[22][8]

YouTube and commentary

Browne's YouTube channel had about 571,000 subscribers in October 2026.^[9] Its videos cover web frameworks, AI models and developer tools. He told Business Insider in 2025 that his favourite thing about AI was that it made things the industry had treated as "super valuable and holy" "incredibly cheap and easy to throw away".^[23]

Journalists have quoted him on disputes in technology. In April 2023 he confirmed that tweets restricted to a user's Twitter Circle could be seen by people outside it, and told TechCrunch that Twitter "seems to be outright failing to filter out private content before serving it to users".^[2] In June 2023 he told Kotaku that Twitch's new branded-content rules would harm its largest streamers and its events most.^[12] In August 2025 he was among the developers in one of OpenAI's launch-day videos for GPT-5; he first praised the model, a week later posted a video titled "I was wrong about GPT-5", and later said that he still considered it "an incredible model".^[11][24] In May 2026 VentureBeat quoted his warning that changes to Anthropic's Claude subscriptions sharply reduced the usage available through third-party tools, including T3 Code.^[25]

Benchmarks and experiments

In May 2025, after Anthropic's system card for Claude 4 described the model contacting authorities when shown evidence of wrongdoing, Browne published SnitchBench, a benchmark that tests how readily AI models report a user to government agencies or the media.^[26] The programmer Simon Willison called it "a great example of an eval" and wrote that it showed the behaviour was not unique to Claude.^[26] In July 2025 Browne reported that Grok 4 contacted the government in every one of his test runs.^[27] Researchers have used SnitchBench as a point of comparison in academic work on AI "whistleblowing".^[28]

In October 2025 Browne published benchmarks comparing the speed of server-side JavaScript on Cloudflare Workers and Vercel; his first results showed Cloudflare slower by a factor of up to 3.5 on some CPU-heavy tasks. Cloudflare responded with its own analysis and changes to its platform, after which it said that it matched Vercel in every case but one, and thanked Browne for "highlighting areas where we could make improvements".^[29]

In October 2026 The Register reported that Browne had used AI coding agents to port Microsoft's TypeScript compiler to Rust. OpenAI models reached about 84 percent compatibility after using more than $400,000 worth of API-equivalent tokens; Anthropic's Claude Opus 5.5 then produced a working version for about $24,000 worth, by porting Microsoft's Go implementation of the compiler, and the result passed the TypeScript conformance tests with nearly 100 percent accuracy. Browne said that the port ran twice as fast as Microsoft's own Go-based rewrite, and that he had not inspected its code.^[10]

References

1. Geoff Weiss, 19 Former Twitch Staffers Who Have Launched Their Own New Startups, Business Insider, 16 May 2022.
2. Amanda Silberling, Twitter Circle tweets are not that private anymore, TechCrunch, 10 April 2023.
3. Nathan Grayson, How Twitch lost its way, The Washington Post, 23 March 2023 (subscription required).
4. T3 Chat, Y Combinator, accessed 15 October 2026.
5. Christopher Mims, Cutting-Edge AI Was Supposed to Get Cheaper. It’s More Expensive Than Ever, The Wall Street Journal, 29 August 2025 (subscription required).
6. Ina Fried and Madison Mills, OpenAI releases GPT-5.6 and ChatGPT Work tool, Axios, 9 July 2026.
7. Amanda Caswell, Developers are secretly hoping OpenAI fails to ship this month, The New Stack, 5 October 2026.
8. Introduction, Create T3 App, accessed 15 October 2026.
9. Theo - t3.gg, YouTube, accessed 15 October 2026.
10. Joab Jackson, $24,000 AI experiment ports TypeScript compiler to Rust, The Register, 15 October 2026.
11. Richard MacManus, GPT-5: A Choose Your Own Adventure for Frontend Developers, The New Stack, 14 August 2025.
12. Alyssa Mercante, Twitch's New Ad Rules Are Very Bad For Streamers, Kotaku, 6 June 2023.
13. Dave Kiss, Mastering Video Uploads and Streaming: Insights from Theo Browne (t3.gg), Mux, 26 November 2024.
14. Theo Browne, LinkedIn, accessed 15 October 2026.
15. pingdotgg/uploadthing, GitHub, accessed 15 October 2026.
16. Ping Labs, ping.gg, accessed 15 October 2026.
17. Release v0.0.2, pingdotgg/t3code, GitHub, 7 March 2026.
18. T3 Code, t3.codes, accessed 16 October 2026.
19. lakebed, npm, accessed 15 October 2026.
20. Lakebed, lakebed.dev, accessed 15 October 2026.
21. Anna Heim, YC alum Adam raises $4.1M to turn viral text-to-3D tool into AI copilot, TechCrunch, 31 October 2025.
22. t3-oss/create-t3-app, GitHub, accessed 15 October 2026.
23. Emma Cosgrove, Developers Say to Stay Ahead in AI, 'Burn the Boats.', Business Insider, 27 April 2025.
24. Richard MacManus, Is GPT-5 a Coding Powerhouse or Maintainability Nightmare?, The New Stack, 6 September 2025.
25. Carl Franzen, Anthropic reinstates OpenClaw and third-party agent usage on Claude subscriptions — with a catch, VentureBeat, 13 May 2026.
26. Simon Willison, How often do LLMs snitch? Recreating Theo's SnitchBench with LLM, Simon Willison's Weblog, 31 May 2025.
27. David Uzondu, Grok 4 will always snitch on you and email the feds if it suspects wrongdoing, report says, Neowin, 12 July 2025.
28. Kushal Agrawal, Frank Xiao, Guido Bergman and Asa Cooper Stickland, Why Do Language Model Agents Whistleblow?, arXiv, 21 November 2025.
29. Kenton Varda, Unpacking Cloudflare Workers CPU Performance Benchmarks, Cloudflare Blog, 14 October 2025.