The API bill is the usual reason teams look at their own hardware, and privacy is the better one. Which benchmarks to trust for agents, how far you can quantize before logic breaks, what a single A100 on GCP costs, and serving Qwen 3.5-27B with vLLM.
WordPress.org now runs an MCP server for the plugin directory, so an AI assistant can validate readme.txt, check where a submission sits in the queue and read the guidelines from the source. The review team still holds you responsible for the code.
AI coding assistants write decent functions and miss the system around them. A naive REST endpoint next to a reviewed one, two outage stories, and the workflow I use to keep an agent on a leash.
Telex is Automattic’s experimental tool for generating block themes from a prompt. What the output actually contains, the templates you have to ask for, and why I test every generated ZIP in Playground first.
Shadow AI shows up because the official workflow is slower than the shortcut. Here is how I read that usage as a signal, and what a governed option looks like inside a WordPress stack.
AI coding assistants write PHP that reads fine and falls over in production. A WooCommerce audit, the naive fetch they keep producing, and the refactor I run it through.
Generating a thousand lines of PHP takes seconds, so review is where the time goes. The review skill I run before opening a PR, why I ask for HTML reports, and the WooCommerce checkout bug that nearly shipped.
I moved a 500-line prompt into a Claude Code Skill and stopped re-pasting it every morning. This covers the three loading stages, how to write a description Claude will act on, and the three implementation patterns.
Building an AI application is mostly plumbing: keys kept out of the repo, a plan for 429 responses, and a chunking engine for anything longer than the token limit. Notes on what keeps an AI feature running once it leaves your laptop.
Nobody trains a vision language model from a blank slate. You freeze a ViT, train a Q-Former to bridge image features into the text embedding space, and wrap the language model in LoRA. This is how those three pieces fit and where the GPU hours go.