Theo - t3․gg
Software dev, AI nerd, TypeScript sympathizer, creator of T3 Chat and the T3 Stack.
I can't believe nobody's done this before...
A deep dive into OpenAI's websocket move, why it matters for maintaining state across tool calls, and how it enables much more efficient and scalable agent interactions. The video also covers practical implications, caching, and standardization, with a sponsor plug and future outlook.
I'm so f***ing tired of Obsidian.
The video explains how the creator runs their content pipeline in Notion, highlights its strengths for team collaboration and flexibility, but also calls out its fragility and costs, before contrasting it with Obsidian and sharing the rationale behind sticking with Notion for their workflow.
Cursor, Claude Code and Codex all have a BIG problem
The video argues that modern AI powered development tools expose a deeper problem: the codebases they operate on are often sloppy and hard to maintain, which AI cannot fix simply by patching the surface. The speaker advocates for strategies that keep codebases clean, plan-driven, and sometimes rebuilt from scratch to sustain long term velocity and reliability.
Software engineering is dead now
The video argues that AI and AI-assisted tools are rapidly transforming software development, driving a shift from large, code-centric teams to smaller, more automated, and intelligence-centered organizations, while also detailing layoffs and strategic pivots at major companies. It mixes personal experience with industry observations and ends with a call for developers to adapt by embracing automation, restructuring teams, and engaging more with users and products.
Anthropic just refused Trump’s order for a lethal Claude model
The video discusses the political and regulatory pressure surrounding Anthropic, including government threats and proposed AI use in national security, while the host defends Anthropic and argues for free markets and responsible use of AI in defense. It surveys the legal tools, safety safeguards, and public responses, concluding with a call to support Anthropic and a proposed open letter from industry employees.
They cut Node.js Memory in half 👀
The video explains how a Node.js memory reduction technique called pointer compression, enabled by a Docker image swap (Node caged), can cut memory usage by about 50% with modest latency impact, driven by real-world testing and detailed V8 internals. It also discusses the engineering effort, historical constraints, and practical implications for production deployments and edge workloads.
My new app is really stupid (I wrote none of the code)
The video introduces a new side project called Quip Slop, a AI driven game inspired by Quiplash that runs across a CLI and a web UI with live Twitch voting, plus a behind the scenes look at building and streaming the project end to end.
Anthropic is lying to us.
The video critiques Anthropic’s distillation claims, breaking down what distillation means, the evidence for and against labs allegedly distilling models, and the practical implications for safety, data use, and enterprise benchmarking. It also analyzes real world examples from labs like DeepSeek, Moonshot, Miniax, and Cursor to illustrate how distillation plays out in practice and why skepticism is warranted.
Delete your CLAUDE.md (and your AGENT.md too)
The video argues that agent MD and Claude MD files are not a guaranteed boon for AI coding tasks and emphasizes context management, practical benchmarks, and when to rely on or discard such files. It also shares experiments, a sponsor segment for Daytona, and actionable guidance on building robust prompts and codebases to steer AI behavior more reliably.
Gemini 3.1 Pro is the smartest model ever made
The video analyzes Gemini 3.1 Pro's benchmark performance relative to peers, explores real world usability versus its vast knowledge, and examines tool use, harness design, and CLI reliability, ending with practical guidance and comparisons to other models.
The Future of TypeScript
The speaker surveys TypeScript's impact, current challenges, and a bold roadmap toward faster, more reliable type checking and closer alignment with Go, while highlighting new features and the evolving ecosystem shaped by large scale use and AI-driven development.
Where Should You Deploy In 2026?
The video offers a practical framework for choosing where to deploy your code, comparing serverless options (especially AWS Lambda and Versell), traditional VPS routes, and edge/cloud providers, and ending with the creator’s top recommended picks and cautions based on cost, DX, and reliability.
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.