When AI's 'Thinking' Mode Actually Makes It Worse
Reasoning models are a genuine leap — on the right problems. On the wrong ones, thinking longer makes AI slower, pricier, and less accurate. Here's the test.
Aug 21, 202610 min read
Command Palette
Search for a command to run...
Series
Building with AI
Plain-English deep dives into how AI actually works under the hood — and how to build with it without the hype. Prompting that gets real results, how RAG systems retrieve before they answer, working with coding agents, and the compounding-error math behind why autonomous agents fail in production. For developers and the technically curious who want the practical layer beneath the buzzwords. No jargon, no hype — just clear mental models you can use.
AI Writes Code That Looks Right. That's Exactly the Danger.
AI-generated code fails differently than human code — it looks polished and confident while being subtly wrong. Here are the 5 failure modes and how to review for them.
Aug 24, 20268 min read
How AI Learned to Actually Do Things: Tool Calling and MCP Explained
A language model can only produce text — so how do AI agents check the weather, query databases, and book meetings? The answer is tool calling. Here's the whole mechanism.
Aug 27, 20269 min read
The AI Security Hole You Can't Patch: Prompt Injection Explained
A stranger can hide a sentence in a webpage or email and your AI will obey it. It's OWASP's #1 AI risk — and unlike SQL injection, it can't be cleanly fixed. Here's why, and how to contain it.
Aug 28, 20269 min read