Theo Browne, known as Theo or t3.gg, is a developer, video creator and founder of T3 Chat. His work includes the T3 Stack, a web-application starter, and products for file uploads, streaming and coding-agent interfaces.[1][2]
Start with Create T3 App →
GitHub 
X Career
Browne’s public profile identifies him as CEO of Ping.gg and founder of T3 Chat. His site connects those products to his YouTube channel and live streams, where he discusses development tools. The same profile lists UploadThing for file uploads, Ping.gg for streamers, and smaller utilities such as QuickPic.[1]
Contributions
A starter with choices built in
Create T3 App scaffolds a TypeScript web application and lets developers choose the pieces they need, including database access, authentication and communication between browser and server. Its documentation explains the T3 Stack’s preference for keeping types consistent across that boundary. The project is developed through the t3-oss community; Browne identifies himself as the stack’s creator.[1][3]
A shared interface for coding agents
T3 Code provides one interface for several coding-agent tools, including Codex and Claude Code, while using the subscriptions developers bring. Its product site links to the open-source implementation. Browne’s profile includes it among his current products; that supports his involvement without treating every feature or line of code as his sole work.[1][4]
Tools, writing & talks
Tools
Start here
Create T3 Appcreate.t3.ggA configurable starter for TypeScript web applications built around Next.js.
Start with the component explanations and choose only what your app needs. Basic JavaScript, TypeScript and React knowledge will make the generated project easier to understand.
Apps
Writing, research & talks
Theo explains the harness layer: the tool definitions, call syntax, parser and result loop that turn an LLM into an agent. It separates harness from interface, shows where MCP fits, and uses Claude Code, Codex and T3 Code as examples.
Catalog workOriginal sourceT3 Code Stable shipped over 1,400 changes. Highlights include beta early compaction of idle Claude threads to preserve cache, adding projects to other machines, PR diffs grouped by code, tests and docs, webhook task filters, and a more experimental Nightly channel.
Catalog workOriginal sourceTheo notes that Muse Wonder Full is a significantly stronger model than the standard Wonder available over API, with Full restricted to subscribers, and compares token efficiency on Artificial Analysis.
Catalog workOriginal sourceArgues that recent leaps in 3D generation come from labs adding 3D training data and RL environments, so impressive 3D demos are poor proxies for a model's real-world coding capability.
Catalog workOriginal sourceTheo challenges a viral agent-spend estimate, showing how cached tokens inflate cost and citing Artificial Analysis data that price at constant intelligence fell about 30x in five months.
Catalog workOriginal sourceTheo announces tsc-rs, an open-source Rust rewrite of the TypeScript compiler, type checker and LSP built entirely by agents over five months, with a spend comparison between Codex and Opus runs.
Catalog workOriginal sourceAfter agents wrote 200+ ad hoc PR-watching scripts, Theo shares a prompt to audit coding-agent history for reinvented babysitting logic, flawed implementations and token waste.
Catalog workOriginal sourceT3 Code cut GitHub rate-limit consumption by over 75% by dropping CLI calls for direct APIs and rotating between GraphQL and REST with fallbacks, making PR monitoring more reliable.
Catalog workOriginal sourceT3 Code shipped in-app visualization, letting agents build dynamic interactive experiences inside the thread using CSS style tokens that match the user's theme.
Catalog workOriginal source- Anthropic Actually Fixed Opusyoutube.com
Theo (t3.gg) reports Opus 5.5 fixed the reliability problems that made Opus 5 unusable for real coding work, now completing a full day's tasks on about 20% of his weekly Claude usage.
Catalog workOriginal source - They lied...youtube.com
A skeptical read of Anthropic's Claude Code limit change, arguing the advertised 25 percent increase translates into roughly 17 percent less usable capacity in practice. Relevant to anyone sizing agent workloads against published quotas.
Catalog workOriginal source - I'm done with terminalsyoutube.com
Theo argues the terminal is the wrong surface for coding with agents, and that after a career spent living in it the workflow now needs something else. A position piece on where the coding-agent interface should go next.
Catalog workOriginal source - Claude watermarks your code nowyoutube.com
Anthropic is watermarking output from Claude models launched on or after 2 August, and this walks through how the marks work and why removing them already looks straightforward.
Catalog workOriginal source - xAI just caught up (Grok 4.6 is here)youtube.com
Grok 4.6 gains five points on the Artificial Analysis Intelligence Index to reach 61, but burns 30% more tokens per run and nearly doubles in price, which raises the question of whether the trade still makes sense.
Catalog workOriginal source - I made Claude smarter by writing it a letteryoutube.com
Theo walks through his full rewrite of his global AGENTS.md and CLAUDE.md, including why his skill descriptions are built around trigger keywords.
Catalog workOriginal source
