-
What Nobody Explains About How a Transformer Actually Works
A precise, jargon-minimal explanation of transformer attention mechanisms for smart readers without an ML background. What attention does and what it doesn't.
-
I Compressed 10 Years of Bookmarks Into One Searchable AI Brain
How to turn a decade of browser bookmarks and Pocket exports into a local semantic search archive. Dead-link auditing script, single-file page archiving, and embedding-based search with AnythingLLM on a Mac.
-
Cline and Roo Code: Roo modes that codify team-specific workflows
A field-tested take on Roo modes that codify team-specific workflows with Cline and Roo Code: what it rewards, where it breaks, and how to keep the workflow honest.
-
Reviews Without Benchmarks: How to Evaluate a Product Based on Real-World Usage
Learn to evaluate products through real-world usage instead of synthetic benchmarks. Practical frameworks for assessing quality based on daily experience.
-
Why the Metaverse Failed and What That Means for AI's Optimistic Scenarios
The metaverse had $100B, real VR hardware, pandemic tailwinds — and failed anyway. The lesson for AI's optimistic scenarios is more relevant than anyone admits.
-
The Freelancer's AI Stack — Run Your Whole Business From a MacBook
A complete AI stack for freelancers on a MacBook. Proposal drafting prompts, invoice automation with Shortcuts, local Whisper meeting transcription, contract review, and bookkeeping with LLM-generated pandas — with real time savings and costs.
-
JetBrains AI Assistant: Writing specs as Markdown that Junie can consume
A field-tested take on writing specs as Markdown that Junie can consume with JetBrains AI Assistant: what it rewards, where it breaks, and how to keep the workflow honest.
-
Why Power Users Are the Most Dangerous Target Audience for a Product
Discover why designing products for power users often leads to failure. Learn how vocal minorities can mislead product strategy and why mainstream.
-
The Grid Problem Nobody in AI Is Talking About Publicly
AI's real constraint isn't chips or talent — it's electrical grid capacity. Grid connection requests take 4-5 years, and the queue has grown 700% since 2020.
-
MCP Explained — Why Your AI Can Finally Use Your Mac Apps
Model Context Protocol (MCP) explained for Mac power users. How to connect Claude Desktop to your filesystem, calendar, and notes with real config examples, plus the security rules you need before installing any MCP server.
-
Windsurf: Pairing Cascade with a linter that refuses bad taste
A field-tested take on pairing Cascade with a linter that refuses bad taste with Windsurf: what it rewards, where it breaks, and how to keep the workflow honest.
