News from January 2026
A deep-dive talk that defines Generative UI, compares static, open-ended (MCP Apps), and declarative (A2UI) patterns, and shows how the AG-UI protocol streams state and UI to build real, interactive agent-native apps.
A practical, opinionated walkthrough comparing Claude Code, OpenCode, Cursor, and GitHub Copilot across models, IDE/TUI workflows, features like agents, skills, memory, and subscriptions, concluding that all are capable and choice depends on models, workflow, and prompts.
Step-by-step Docker stack to run Paperless-ngx, optionally add local AI with Ollama and Open WebUI, and dramatically improve OCR and metadata using Paperless-AI and Paperless-GPT for private, searchable documents.
A hands-on deep dive into Clawdbot showing setup, capabilities, real use cases, costs, and serious security cautions—why it feels like a powerful, slightly scary glimpse of where AI is heading.
Practical walkthrough of TanStack AI showing how to build a chat app with server and client tools, adapters for multiple AI providers, streaming via SSE, and typed tool calls for agentic workflows.
A practical comparison of Apple M-series Macs and Nvidia GPU setups for running local LLMs at home, covering costs, performance, power, and when each option makes sense.
Explains Anthropic’s new MCP Tool Search that defers loading tool definitions to cut context-token usage, outlines regex vs BM25 search modes, best practices for concise tool descriptions and server instructions, and a 4-step client implementation with pitfalls to avoid.
Overview of Google’s open-source Universal Commerce Protocol (UCP), showing how it standardizes secure, extensible agentic commerce with a demo flower shop workflow and Google’s reference implementation.
A quick demo showing how to build and use Claude Code skills with progressive disclosure to run FFmpeg scripts for video downscaling.
Overview of A2UI, an open protocol where AI agents output declarative JSON "recipes" that your app renders with approved components, enabling safe, runtime UI generation without code in production and contrasting it with code generators and prebuilt widgets.
Theo shares how using Claude Code and Opus 4.5 radically accelerated building web and mobile projects, restructured repos, and even automated system tweaks—changing how he codes and works.
A reflection on rapidly evolving AI-assisted coding—arguing that while AI now produces most lines and accelerates delivery, developers shift toward specification, orchestration, and rigorous review, with new challenges for juniors and quality tradeoffs.
A practical workflow tip: plan complex tasks in a large chat, then execute each task in a fresh chat to avoid context-window noise and reduce hallucinations and token cost.