WordPress 7.0 puts WordPress Core AI in the codebase, and not behind an experimental flag this time. I have been building on this platform since the 4.x days, and I have watched plenty of announced innovations settle into legacy bloat. This one reads differently to me, because the work is going into the plumbing instead of into another panel in the editor.
The Core AI Team is heading to WordCamp Asia 2026 in Mumbai and spending Contributor Day, April 9, on those tools. For a developer that means hands on the provider-agnostic AI Client and on the discovery logic behind the Abilities API.
Why WordPress Core AI matters
Until now, AI in WordPress meant writing your own cURL calls to OpenAI or Anthropic. Every plugin shipped its own implementation, which piled up technical debt and caused race conditions whenever two of them hooked into the same data. WordPress Core AI replaces all of that with one library everyone shares.
The wp_ai_client_prompt() API is the centerpiece. It hides the provider, so your plugin calls one function and the site owner picks Google, Anthropic, or a local model from a single interface. The Abilities API handles the other half: a central registry of what the site can actually do, so an AI agent can discover that functionality safely.
I went through the details in Essential WordPress 7.0 AI Client, and the building blocks approach looks like the only version of this that scales without breaking the ecosystem.
A closer look at the AI Client API
Instead of juggling raw API keys and endpoint headers, you get a standard prompt flow. Here is a stripped-down version of a generation request in 7.0.
<?php
/**
* Example of using the native WordPress 7.0 AI Client
* Prefixing with bbioon_ as per senior dev best practices.
*/
function bbioon_generate_seo_meta( $post_id ) {
$content = get_post_field( 'post_content', $post_id );
// The new provider-agnostic prompt API
$response = wp_ai_client_prompt( "Summarize this for a meta description: " . $content )
->using_temperature( 0.5 )
->generate_text();
if ( is_wp_error( $response ) ) {
error_log( 'WordPress Core AI Error: ' . $response->get_error_message() );
return;
}
update_post_meta( $post_id, '_yoast_wpseo_metadesc', wp_kses_post( $response ) );
}Notice the error handling. Check is_wp_error() every single time. External APIs go down and transients expire, and if you skip the check, the first sign of trouble is a client calling you about a broken checkout or a white screen of death.
Join the Core AI table in Mumbai
If you are attending WCAsia 2026, the Contributor Day schedule at the Jio World Convention Centre is packed, and the Core AI table is a working session where people write code. Contributors including David Levine and Aslam Doctor are leading it, and the focus is on:
- Testing the MCP Adapter, which bridges WordPress to assistants like Claude over the Model Context Protocol.
- Refining the Abilities API so custom plugin logic is discoverable without exposing sensitive data.
- Benchmarking with
wp-benchto see how different models handle WordPress-specific tasks.
Our earlier post on Solving the Inversion Error in Safe AI Design covers the security problems sitting behind that work.
Getting ready for the 7.0 release
You do not have to wait for the general release. Set up a local site, install the WordPress Beta Tester plugin, and switch it to the Bleeding Edge channel. Add the AI Experiments plugin on top of that and you can poke at the WordPress Core AI building blocks before any of it is finalized.
If the WordPress Core AI work is eating your dev hours, I can take it off your plate. I have been wrestling with WordPress since the 4.x days, so I know where the bottlenecks usually sit.
What this means for developers
The 7.0 release pushes WordPress toward an agentic ecosystem, one where the CMS has some grasp of its own data. That is a change about context rather than about blocks. Whether you contribute in person in Mumbai or remotely over Slack, the decisions made on the core building blocks now are what the platform will be stuck with for years. Better to adjust your code while the API is still moving than to pay for it in support tickets later.
