The WordPress AI Experiments plugin has a new release, and it reads like Core is done with the chatbot novelty phase. Plenty of the ecosystem is still wrapping fetch() calls to OpenAI inside theme functions. The Core AI team is building architecture instead, and 0.2.0 is the first version that feels like a foundation rather than a demo.
If you read my take on integrating the WordPress AI client, you know where I land on stability versus hype. This release ships editorial tools and, for developers, the Abilities Explorer. Between them they answer how you manage AI-powered actions without turning functions.php into a prompt-injection surface.
AI-powered excerpts, with a review step
The headline feature in WordPress AI Experiments 0.2.0 is excerpt generation. Excerpts are the field everyone skips, left blank so WordPress falls back to the first 55 words of the post. This experiment lets an editor generate a plain summary from the sidebar instead.
The naive version of this would be a button that overwrites the textarea. The team built a reviewable workflow: the author generates, then edits or throws it away. The human stays the last filter, which is the pattern worth copying in client work.
The Abilities Explorer is a debugger for AI
If you build custom blocks or extend the editor, the Abilities Explorer is the part to look at. It is an admin screen under Tools > Abilities Explorer that lists every AI ability registered on the site and lets you run inputs through them in the browser.
Debugging how the official plugin handled a given prompt used to be tedious work. Now you can confirm your custom ability registered correctly and read back exactly what the model returned. For anyone extending WordPress AI Experiments, that visibility matters more than the excerpt feature does.
<?php
/**
* Example: How to register a custom Ability in WordPress AI Experiments.
* Avoid hardcoding API calls; use the registered abilities framework.
*/
function bbioon_register_technical_summary_ability() {
if ( ! function_exists( 'wp_ai_register_ability' ) ) {
return;
}
wp_ai_register_ability( 'bbioon_tech_summarizer', [
'label' => __( 'Technical Summary', 'text-domain' ),
'description' => __( 'Converts complex code descriptions into readable summaries.', 'text-domain' ),
'prompt' => 'Explain the following code logic for a non-technical project manager:',
'capability' => 'edit_posts',
] );
}
add_action( 'init', 'bbioon_register_technical_summary_ability' );Architecture and performance
Underneath, the framework is being moved to TypeScript with the modern @wordpress/build tooling, and settings are heading toward centralized settings built on DataViews. The release also sets up what is coming next: alt text generation and summarization for long-form posts.
If keeping up with WordPress AI Experiments is eating your dev hours, that is the kind of work I take on. I have been building on WordPress since the 4.x days.
Standardizing the AI interface
The practical advice is to stop hand-rolling a one-off AI integration for every site. The WordPress AI Experiments plugin is working out the WordPress way to talk to models. Build on the Abilities API and your code keeps working whichever LLM the user connects, whether that is OpenAI, Claude, or a local Llama through Ollama. Your themes and plugins stay portable.
