AI tools, apps, and workflows — what works, what does not, how to pick.
An AI coding agent without version control is a power tool with no off switch. Its built-in undo expires and misses things — Anthropic itself says checkpoints are ‘not a replacement for git.’ Here is why git and GitHub are the safety net every LLM coding agent assumes you have: diffs to see what it did, branches to contain it, pull requests to review it, and @claude to put the agent right inside GitHub.
If you describe an idea in a chat window and then copy-paste the code into your editor, you are doing the agent’s job by hand. Here is the modern workflow: brainstorm in Claude.ai, export the spec as a Markdown file, hand it to Claude Code on Opus 4.8, and run it safely in Docker — no couriering snippets, no errors copied back and forth.
The exact playbook for letting Claude Code build and configure your WordPress site end-to-end — Application Passwords, the REST API, sample prompts from a real build, and how to use Chrome MCP when the API hits its limits.