The future of WordPress depends on defining what artificial intelligence will achieve in WordPress, for developers, editors, and project leadership alike.
Remember that you can listen to this program from Pocket Casts, Spotify, and Apple Podcasts or subscribe to the feed directly.
Program transcript
Hello, I’m Alicia Ireland, and you’re listening to WPpodcast, bringing the weekly news from the WordPress Community.
In this episode, you’ll find the information from February 2 to 8, 2026.
This week, Matt Mullenweg, co founder of WordPress, published an editorial titled Some Provocative AI Thoughts, in which he opens a broad reflection on how artificial intelligence could redefine the future of WordPress. It is not a technical proposal or a roadmap, but an invitation to think long term about the impact of AI on how software is built, how people contribute to the project, and how decisions are made. Matt suggests that AI will not just be a supporting tool, but an actor that can change fundamental concepts such as authorship, contribution, and the pace at which the project itself evolves.
The text raises uncomfortable but necessary questions, such as what it means to contribute when much of the work may be mediated by AI, how trust and human judgment are maintained in an increasingly automated environment, or what risks emerge if these capabilities become concentrated in a few hands. Rather than offering answers, the editorial seeks to spark an early conversation within the community, with the idea that ignoring these debates or addressing them too late could negatively shape the future of WordPress, while tackling them now allows boundaries, values, and direction to be defined more clearly.
A new piece of documentation titled AI Guidelines for WordPress establishes a shared framework for the development and integration of artificial intelligence features within the WordPress ecosystem, with the goal of ensuring consistency, security, and alignment with the project’s values.
The guidelines make it clear that AI in WordPress must be optional, extensible, and non intrusive, avoiding unexpected automated behavior and ensuring that a site continues to function correctly even if no AI provider is configured. They also emphasize that core should not include models, credentials, or default providers, and that these must be handled through plugins.
The document defines key principles such as transparency, user control, privacy, and portability, recommending provider abstraction, avoiding rigid dependencies, and designing APIs that can evolve without breaking compatibility. It stresses that AI powered user experiences should be built on well defined layers such as Abilities and the WP AI Client, clearly separating infrastructure from product.
The most important points of the WordPress AI Guidelines are:
- You are responsible for what you submit. AI can help, but it is not a contributor. You must understand every line and ensure it is correct, secure, and appropriate for the project.
- Transparency. If AI was used in a meaningful way, it should be disclosed in the pull request or contribution.
- Licensing. Only use tools and outputs that can be licensed as GPL 2 or later. If you are not sure, do not submit it and ask in the appropriate channels.
- Quality over volume. Avoid bloated or generic content, what they refer to as no AI slop.
Regarding the use of AI for code:
- Scaffolding, boilerplate, suggested refactors, as well as draft comments and documentation, are acceptable, always with human review.
- Avoid changes that are not understood, patterns or dependencies outside WordPress standards, and copied code with unclear or GPL incompatible origins.
- Before submitting, read everything and remove anything you cannot explain, align with standards, add or adjust tests, run the relevant test suites, and document the commands used.
For tests, documentation, issues, and reviews:
- Tests and QA. AI can be used to propose edge cases and turn manual steps into automation, but tests must be validated to ensure they are realistic, deterministic, and not dependent on unstable external services.
- Documentation and communication. Review function names, hooks, and options, check links, avoid invented references and unnecessary verbosity, and ensure that any materials or resources are also GPL compatible.
- Issues and support. Drafting with AI is acceptable, but issues must be reproduced on real installations and include concrete steps, logs, or screenshots, not generic templates.
- Maintainers. They may request clarification, additional tests, or explanations, and can reject contributions with unclear origin, low quality, or licensing risks. AI cannot be the sole reviewer.
Overall, the guidance calls for applying common sense and avoiding indiscriminate use of AI, treating it instead as a support tool.
A new version of the experimental plugin Gutenberg 22.5 has been released, bringing several notable updates. These include support for per block custom CSS, allowing CSS to be applied to individual block instances without affecting others of the same type, with an automatic has-custom-css class to make style management easier.
Aspect ratio controls have also been added to the Image block, so when using wide or full alignment, the editor displays aspect ratio settings to lock consistent image proportions across different layouts.
In another area, improvements to the List View now display full titles and real content instead of generic labels, making it easier to navigate and structure complex content.
Additionally, focal point controls have been added for fixed backgrounds in the Cover block, a text column option is now available for the Paragraph block, navigation menus can be configured with always open submenus, and initial work on revisions inside the editor has begun.
The Core team has proposed merging the WP AI Client into WordPress 7.0 as development infrastructure that allows WordPress code to communicate with generative AI models through a unified API, integrating HTTP transport, credential handling, REST, JavaScript, and prompt generation, and natively linking it with the Abilities API already in Core.
The goal is to centralize and unify the foundation so that plugins and code can detect whether a model is configured, invoke AI capabilities, and, if none is available, disable AI features without breaking anything. It does not include providers or enable AI by default, nor does it introduce assistant interfaces or product level experiences.
The proposal defines clear scope and boundaries. It aims to provide a WordPress adapted PHP API and Prompt Builder, provider and model abstraction, integration with the wp admin credential system, REST and JavaScript endpoints for editor and admin use cases, and fine grained control over prompt execution, with a strong emphasis on security, privacy, and minimal impact when unused. Credentials, models, and assistant interfaces are explicitly out of scope for Core, and those experiences are expected to live in plugins such as AI Experiments.
The Core AI team has issued an open call to the community to test new AI experiments within the AI Experiments plugin, with the goal of gathering early feedback before making decisions about their future direction. The focus is on exploring new interface experiences and interaction flows with AI assistants, especially in the editor, to understand which approaches work well, which create friction, and which patterns should be discarded early.
Among the new experiments being tested are several approaches to AI assistant interfaces within the editor, aimed at understanding how and when it makes sense to interact with AI during content creation.
These include contextual panels that react to selection or content state, more guided conversational flows for specific tasks such as summarizing, rewriting, or generating drafts, and interface patterns that clearly separate AI suggestions from the user’s final action.
On the Developer Blog, a post titled From Abilities to AI Agents: Introducing the WordPress MCP Adapter has been published, introducing the WordPress MCP Adapter. This is a key piece of the AI Building Blocks initiative that connects the WordPress Abilities API with the open standard Model Context Protocol, MCP, enabling structured and secure interaction between AI agents and WordPress sites.
The adapter supports HTTP and STDIO transport, provides granular permission control and validation, and exposes WordPress capabilities to AI agents without requiring custom integrations for each provider or model. With this architecture, an agent can, for example, list, create, or modify content on a site while respecting user permissions and approvals, and WordPress can even act as an MCP server or MCP client to interact with other AI driven services or workflows.
The Test team training program, after a month of intensive work, concluded with a positive assessment and clear lessons for improving contributor onboarding. It started with nine participants and finished with six, who during the process worked on testing, documentation improvements, meeting facilitation, and continuous feedback in a dedicated Slack channel. The structure evolved over time. Week one focused on testing protocols, week two on meeting processes and feedback collection, week three on documentation improvements and the start of a video program, and week four on closing open questions and reviewing the team’s future goals.
In terms of results, the biggest impact was on the needs testing ticket queue, which dropped from 487 to 264, nearly a 50 percent reduction, showing that the protocol already enables a high and consistent pace. Documentation improvement proposals were also collected in a GitHub repository, and it was announced that the video based Test Contributor Pathway is nearly complete, with plans to open a beta program to gather feedback before its official launch. The program is described as highly effective but exhausting to run, and repeating it will require making it more sustainable and better distributing the workload among new members as they join.
And finally, this podcast is distributed under a Creative Commons license as a derivative version of the podcast in Spanish; you can find all the links for more information, and the podcast in other languages, at WPpodcast .org.
Thanks for listening, and until the next episode!
Leave a Reply