Development
WordPress Is Becoming an Agent Platform, Not Just an AI Writing Tool
The easiest way to misunderstand WordPress’s AI strategy is to focus on content generation.
Yes, WordPress can suggest the next sentence, summarize an article, generate images, write excerpts, and help moderate comments. Those are useful features. They are also the least interesting part of what is happening.
The bigger shift is underneath them.
WordPress is developing a standardized way for AI models to understand what a website can do, connect to an appropriate provider, follow WordPress permissions, and perform actual work. The chatbot is the showroom. The plumbing is the product.
Over the next 6 to 12 months, between roughly February and August 2027, that infrastructure will begin changing how plugins are built, how stores are operated, and what clients expect from WordPress professionals.
WordPress 7.0 Quietly Changed the Baseline
WordPress 7.0 “Armstrong” introduced a provider-neutral AI Client, a central Connectors screen, and deeper integration with the Abilities API. Plugins no longer need completely separate integrations for OpenAI, Anthropic, Google, or another provider. They can build against a common WordPress interface while site owners choose the service and credentials they prefer.
That might sound like developer plumbing because it is developer plumbing. But WordPress’s biggest successes have usually come from boring infrastructure that allowed thousands of developers to build less-boring products.
The REST API did not look like a website builder. Blocks did not initially look like a complete publishing system. Likewise, the AI Client is not supposed to be the final AI product. It is the common foundation on which competing products can be built.
My prediction is that Core will continue resisting the temptation to ship one giant built-in assistant. The WordPress AI team has said that its focus for the AI Client remains on empowering plugin developers. Planned improvements include streaming and embeddings, while experimental user features continue to live in the canonical AI plugin. The details are outlined in the team’s preview of what is coming to the AI Client.
That is probably the correct decision. A universal WordPress chatbot would become outdated almost immediately. A stable platform that lets developers use changing models is much harder to replace.
Agents Will Move From Generating Content to Operating Websites
Most current WordPress AI products wait for a user to press a button. The next generation will increasingly perform multi-step tasks.
The Abilities API lets WordPress and plugins register specific operations with typed inputs, outputs, and permission checks. The official WordPress MCP Adapter can then expose approved abilities to compatible AI agents. An agent might inspect a site, find a problem, call an authorized ability, verify the result, and produce a report.
This is a much bigger market than paragraph generation.
A membership plugin could expose abilities for importing members, diagnosing failed payments, creating courses, checking configuration, or generating reports. An agency could connect those abilities to maintenance workflows. A site owner could ask an assistant to investigate why registrations dropped without teaching it the location of twelve different settings screens.
Within twelve months, I expect “What abilities does your plugin expose?” to become a legitimate product question for larger WordPress plugins.
Admin interfaces will not disappear. Humans still need dashboards for configuration, oversight, and exceptional cases. But routine operations may increasingly happen through agents rather than menu archaeology.
Site Context Will Matter More Than the Model
Generic AI produces generic output. Every business eventually discovers this, usually after publishing something that sounds like it was written by an enthusiastic kitchen appliance.
The proposed Knowledge and Guidelines system may therefore be more important than another writing feature. The proposal would let sites store voice, tone, editorial standards, image preferences, and block-level rules as native WordPress content with revisions and permissions. Both human authors and AI agents could use that information.
This turns WordPress into the owner of its own context.
Instead of repeatedly pasting brand instructions into different services, a site could maintain one authoritative set of guidelines. Plugins, editorial assistants, translation tools, and automation systems could all work from the same source.
I expect this to become a competitive battleground. The winning AI products will not necessarily use the cleverest model. They will retrieve the right WordPress content, understand the site’s history, and reliably follow its rules.
Plugin Development Will Become Faster and Much Noisier
The new WordPress Playground MCP integration allows coding agents to read WordPress files, run PHP, manage the site, and test changes inside an isolated browser-based installation. This creates a repeatable environment where agents can generate code and immediately check whether it actually works.
The WordPress.org Plugin Directory MCP server goes even further. It gives compatible AI tools access to plugin guidelines, readme validation, submission status, reviewer feedback, and the plugin submission process itself. Human review still applies, regardless of how the code was created.
That dramatically shortens the path from idea to functional plugin.
It also guarantees an incoming avalanche of plugins.
The Plugins Team reported that weekly submissions exceeded 500 by March 2026 and reached around 700 per week in May. The review queue briefly climbed to roughly 1,050 plugins before additional reviewers and AI-assisted tooling helped bring it back down. These numbers are detailed in the team’s June 2026 status report.
Code generation will make creating a plugin easier. It will not make understanding backward compatibility, upgrade paths, security boundaries, support obligations, or actual customer needs easier.
The valuable developers will not be those who can make an agent produce 3,000 lines of PHP before lunch. Almost anyone will be able to do that. The valuable developers will know which 2,500 lines should be deleted before dinner.
Expect stronger automated checks, security scanning, provenance discussions, and pressure for clearer maintenance signals in the directory. Reviews, reputation, and long-term support will matter more as code itself becomes cheaper.
WooCommerce Will Be the First Major Proving Ground
WooCommerce already demonstrates what agent-operated WordPress can look like.
The company’s guide to WooCommerce MCP and built-in Abilities shows an AI assistant finding low-stock products, creating products, and interacting with store data using WordPress permissions. MCP support first appeared as a beta in WooCommerce 10.3 and continued maturing through WooCommerce 10.7.
Commerce provides measurable outcomes, structured data, and repetitive workflows. That makes it ideal for operational agents.
The first successful WooCommerce agents will probably not be autonomous shopping geniuses negotiating with other robots. Reality remains disappointingly less cinematic.
They will handle practical work such as:
- Finding low-stock or poorly described products
- Preparing product updates for approval
- Summarizing unusual orders
- Identifying refund or fulfilment patterns
- Generating support context
- Checking store configuration
- Coordinating promotions across several tools
Fully autonomous transactions will move more slowly because payments, refunds, taxes, and customer data carry real consequences. But read-only analysis and approval-based actions should grow quickly.
Security Will Decide How Fast This Actually Happens
An AI agent connected to WordPress acts through a user account and its capabilities.
The WordPress MCP documentation recommends using minimum permissions, dedicated users, read-only abilities where possible, and careful logging of agent activity. Abilities must also be explicitly made available through MCP rather than automatically exposing every WordPress function.
Those recommendations will soon become product features.
I expect hosting companies and security plugins to introduce agent activity logs, temporary credentials, spending limits, approval queues, and alerts for unusual AI-triggered actions.
The first widely discussed WordPress AI security incident may not involve a model becoming sentient and declaring war on the Media Library. It will more likely involve an overly broad permission callback, exposed credentials, or a plugin allowing an agent to perform a destructive action without sufficient confirmation.
That will briefly produce dramatic headlines, followed by better defaults. WordPress has been through this cycle with almost every major expansion of its attack surface.
Agencies Will Sell Fewer Builds and More Systems
AI will continue lowering the cost of basic WordPress implementation. Simple brochure sites, minor customizations, and routine content work will become harder to price as premium services.
That does not eliminate agencies. It changes what clients will pay them to do.
The stronger opportunities will involve designing workflows, connecting business systems, preparing organizational knowledge, governing AI access, reviewing generated code, and maintaining security. Agencies that understand both WordPress internals and the client’s actual operations will have a considerable advantage.
The pitch will move from “We build websites” toward “We make your WordPress operation faster, safer, and partially automated.”
Smaller teams will complete more work. Generalist agencies will feel pressure. Specialists with reliable processes may become significantly more profitable.
WordPress Will Become Easier to Experiment With
AI development works best when agents have a safe, disposable environment in which to test ideas.
My.WordPress.net runs WordPress persistently inside the browser without requiring a hosting plan, domain, or traditional installation. It is built on WordPress Playground and gives users a private WordPress environment that remains available between sessions.
Combined with Playground’s agent tooling, this lowers the cost of experimentation considerably.
A developer could ask an agent to create a plugin, test it inside Playground, validate its documentation, and prepare it for submission without manually assembling a local development stack.
That is powerful. It is also slightly terrifying in the same way that giving everyone a power drill is useful until somebody starts renovating the load-bearing walls.
What I Expect by Mid-2027
By August 2027, I expect several changes to be visible.
Major commercial plugins will publish Abilities or MCP roadmaps. Managed hosts will offer curated AI connectors with usage controls. Approval screens and agent audit logs will become common. Site Guidelines will increasingly supply context across multiple AI features. WooCommerce assistants will perform routine store operations. Plugin-directory debates around disclosure, quality, and provenance will intensify.
What I do not expect is WordPress Core replacing wp-admin with one enormous chat window.
WordPress is taking a more durable route. It is becoming a permission-aware operating layer where humans, plugins, automation systems, and AI agents can work with the same content and capabilities.
That approach lacks the instant demo magic of “Build my entire website from one sentence.”
It may also be far more important.
The next chapter of WordPress and AI will not be about machines generating more words. It will be about giving machines carefully limited ways to do useful work while making sure humans remain firmly in charge.