Notes from cleaning up after AI coding agents: the code compiles fine and still rots. Divergent change and speculative generality are the two smells I keep finding, with a refactored ML pipeline showing the fix.
OpenClaw keeps Claude Code running as a long-lived agent instead of a chat window. My setup puts it in a Docker container, links a Claude Code token, and uses explicit skills for jobs like GitHub PR reviews and email triage.
Data scientist onboarding takes more than Snowflake keys. What the first 90 days should cover: context from the people around you, a working environment by day 5, and a starter project that earns trust.
Tmux keeps your sessions alive when the connection drops. Sessions, windows and panes, the prefix shortcuts worth memorizing, a starter .tmux.conf, and why agents like Claude Code run on top of it.
Nine things I check when a WordPress site is not turning up in AI search: answer-first writing, topical clusters, JSON-LD schema, explicit authorship, and a robots.txt that is not quietly blocking GPTBot.
Linear RAG breaks down as soon as the solution path is not known up front. LangGraph swaps the chain for a stateful graph: a shared state object, nodes that update it, and conditional edges that pick the next step at runtime.
AI tools like Cursor and Claude Code prototype fast, then leave race conditions and unbounded queries behind. A look at what vibe coded WordPress code misses, with a before and after refactor of a WooCommerce loop.
Automattic shipped a developer preview of a Claude Cowork plugin and a set of Skills that build block themes in WordPress Studio. Here is what the workflow does, where it breaks, and why the theme.json output matters.
How I use Claude Code with WordPress Studio to scaffold plugins: setting up a local site and terminal, installing the CLI, the context a prompt needs, and why the naive admin notice it hands you is an XSS bug waiting to happen.
Agentic AI features need more than a good prompt. These are the patterns I use to keep users in control: intent previews, an autonomy dial, an action audit log with a working undo, and escalation when the model is not confident.