Theo - t3․gg
Software dev, AI nerd, TypeScript sympathizer, creator of T3 Chat and the T3 Stack.
I think every company should open source their code.
The video argues for embracing open source as a strategic business move, outlining the risks of closed systems and presenting building blocks and modular approaches as the future of software, driven by user customization and agent-enabled automation.
Anthropic thinks they're Apple. They're actually hypocrites.
The video critiques Anthropic for what the creator calls hypocritical and restrictive practices around iMessage and API access, drawing in comparisons to Apple’s ecosystem and arguing users should have freer, wallet-friendly options. It also delves into how cloud code and Open Code ecosystems differ in openness and how sponsorships and policy enforcement shape developer choices.
How does Claude Code *actually* work?
The video explains what a harness is in AI coding assistants, how it provides tools and context to the model, and why this layers controls (tools, prompts, and history) dramatically impacts performance and behavior. It also compares different harness approaches, demonstrates building one, and discusses practical implications for using tools with models like Claude, Opus, and Gemini.
I’m scared about the future of security
The video warns that AI driven vulnerability research is accelerating, raising existential fears for security and open source, while sharing Defcon anecdotes and personal moments that illustrate how close we are to a breakthrough in exploit discovery and defense. It also discusses regulatory and practical implications for the industry as AI joins security work.
I HAVE A PODCAST NOW
The video is a candid riff about starting a podcast while diving into recent AI industry drama, especially Anthropic's missteps with Cloud Code and subscriptions, and how OpenAI and other tools compare. It also explores various coding agent harnesses like Pi and Codeex, emphasizing practical use, open source options, and the politics shaping AI tooling.
Claude Mythos and the end of software
The video delves into Claude Mythos preview, its restrictive access, and the profound security and societal implications of such powerful AI models, including ongoing defensive initiatives like Project Glass Wing. It also analyzes capabilities, alignment concerns, pricing, and the risk of centralization in AI development.
Bash is bad for agents
The video argues that Bash alone is not enough for AI agents to safely and efficiently interact with code and systems, and explores moving toward safer execution layers, typed environments, and code-driven tool calls to enable scalable, portable AI workflows.
Claude Code is unusable now
The video argues that Claude Code and OpenClaw have become practically unusable for the creator, due to system prompts, aggressive usage controls, and API level restrictions, and he pivots to recommending alternative tools like Codex-based CLI options and T3 Code. He also vents frustration, cites concrete examples, and urges viewers to explore alternatives while highlighting ongoing support from other teams.
I’m serious.
The video argues that closed source software is increasingly problematic in today's AI era and makes a case for open source as a better, more trustworthy model for software development, giving real-life examples from my own projects and industry trends.
We need to talk about the Claude Code rate limits
The video analyzes Anthropic's new rate limits for Claude Code, arguing the change reflects a broader compute crunch, internal GPU allocation pressures, and a mismanaged communication strategy that damaged user trust. It compares enterprise vs subscription economics, discusses how GPUs are scarce and shared, and critiques Anthropic's handling and transparency while noting broader industry dynamics.
I got DMCA'd by Anthropic (not a joke)
The video analyzes a DMCA notice related to Anthropic's Claude Code leak, explains what was affected, and argues that miscommunication and process failures escalated the issue. It ends with a strong critique of keeping Claude Code closed source and calls for open sourcing to avoid similar crises.
BREAKING: Claude Code source leaked
The video analyzes the Claude Code leak, explains source maps and obfuscated JavaScript, and critiques Anthropic's handling and communications, while highlighting potential open source paths and practical lessons for developers and users.
Get daily AI recaps from
Theo - t3․gg in your inbox
Get AI-powered summaries delivered to your inbox. Save hours every week while staying fully informed.