WordPress 6.9 puts the first usable pieces of the WordPress Core AI initiative into the software we all run. Six months of the working group refining these “building blocks” has left the architecture stable enough to build on rather than poke at. If your site is held together with a custom OpenAI integration and half a dozen SDKs, this is the release to refactor against.
I have watched plenty of “core features” arrive with fanfare and then quietly move to the plugin repository. The Abilities API reads differently. It is a functional core whose job is to describe what an AI system is allowed to do inside your site. Instead of every plugin building its own way to “summarize a post,” there is one registry.
The Abilities API: the functional core of WordPress Core AI
The Abilities API is slated for WordPress 6.9 and it is the biggest milestone so far. Treat it as a schema for AI-powered actions, a shared vocabulary that lets internal WordPress systems, and external agents, work out what a site can do. That gives single-shot features like title generation or post summarization a stable contract to sit on, which prompt engineering alone never provided.
Plugin authors will register their AI features as “abilities,” which is what makes them discoverable to the upcoming Command Palette and other workflow systems. The snippet below sketches how that registration looks for a specific task, generating WooCommerce meta data.
/**
* Registering a custom AI ability for product metadata.
* Note: This is based on the current Core AI implementation patterns.
*/
function bbioon_register_meta_ability() {
if ( ! function_exists( 'wp_ai_register_ability' ) ) {
return;
}
wp_ai_register_ability(
'bbioon/generate-product-meta',
array(
'description' => __( 'Generates optimized SEO meta descriptions for products.', 'bbioon' ),
'parameters' => array(
'product_id' => array(
'type' => 'integer',
'required' => true,
),
),
'callback' => 'bbioon_handle_meta_generation',
)
);
}
add_action( 'init', 'bbioon_register_meta_ability' );
function bbioon_handle_meta_generation( $params ) {
// Implementation logic for fetching product data and calling the AI client
return array( 'meta_description' => 'Optimized description text...' );
}WP Client AI: standardizing the provider interface
Provider lock-in has been the worst bottleneck in WordPress Core AI work so far. Moving from OpenAI to Anthropic usually meant rewriting the whole communication layer. The new WP Client AI API (v0.1.0) fixes that with a canonical PHP SDK sitting between WordPress and the LLM providers.
DreamHost and WordPress.com are already building providers against that interface. For a developer it means writing the feature once and having it work whether the user is on a host that supplies its own AI or bringing an API key of their own. There is more on how this fits the longer roadmap for Core 7.0.
MCP adapter: connecting to the external ecosystem
The MCP Adapter (v0.3.0) is the strangest part of this update. Through the Model Context Protocol, any WordPress site becomes an MCP server, so AI clients such as Cursor or Claude Code can talk to WordPress directly. Registered abilities show up on the other side as tools the AI can run.
The bigger change comes in the next phase, when the adapter goes bidirectional. WordPress will be an MCP client as well as a target, consuming external tools and surfacing them in the admin dashboard through the Abilities API. None of that connection existed six months ago.
If WordPress Core AI is eating your dev hours, hand it to me. I have been wrestling with WordPress since the 4.x days and I have watched a few of these shifts go by. Wiring these APIs up properly now is what saves you a painful refactor when 7.0 arrives.
Build on the standards now
The WordPress Core AI roadmap is not complicated: 6.9 brings the server-side Abilities API, and 7.0 will most likely merge the client implementation and the full WP Client AI SDK. Siloed, proprietary AI code inside plugins has a short shelf life from here. If you build sites or run an agency, this is a good moment to spend an evening with the AI Experiment plugin and line your own projects up with the core standards, rather than finding out at 7.0 that your architecture is the legacy one.
