74 essays
An AI coding agent deleted an entire production database in nine seconds. The failure was permissions, not a rogue AI, and least privilege is the fix.
Consumer AI is tuned to flatter you. Here's how to retune your assistants to argue back — system prompts, honest evals and an assistant-adviser split.
Speaking is three times faster than typing, and GPT-Live's full-duplex voice finally keeps up — how to make voice input your default way to prompt AI.
AI can ace one task and fail the next. Here's why the interview heuristic misleads you, and how to map your jagged frontier before trusting the tools.
Prompt engineering had its moment. Context engineering — giving AI the right information rather than cleverer wording — is the skill that matters now.
Menus did two jobs and we only ever complained about one. As AI removes them, discoverability stops being a layout problem and becomes a teaching one.
NotebookLM is now Gemini Notebook, and every notebook now gets a cloud computer that runs code on your documents. What changed and what to test first.
Your AI agent forgets everything the moment a session ends. Here are the four kinds of memory it actually needs, and the practical ways to build each.
Google logged a 32% rise in prompt-injection attacks. If your AI agent reads the open web, it's in hostile territory now, and here are three controls.
Agents now build whole tasks from your description, so your brief becomes the real work. What Notion's spec-driven workflow teaches about briefing AI.