WordPress.com lets AI agents write and publish content with human approval

WordPress.com has expanded its MCP integration and given AI agents not only the ability to read the site, but also to make changes. Now through a regular dialogue, you can ask an agent to prepare a post, assemble a page, restructure materials, and manage comments — with mandatory human confirmation at each step.

What Changed

Automattic added so-called write capabilities to WordPress.com's MCP integration — the right not just to request data from the site, but to perform actions on behalf of the user. If previously the AI-agent connection was mainly useful for reading content, searching materials, and preparing suggestions, now the scenario is changing: the agent can create drafts, assemble new pages, edit existing texts, and manage comments through natural language.

In practice, this looks like a regular conversation with an assistant. The user formulates a task in words, and the agent translates it into actions within WordPress.com.

The important point is that Automattic does not remove the human from the process. Before publishing, changing structure, or other sensitive operations, explicit human confirmation is required. This approach makes the tool closer to an editorial assistant that can work directly in the CMS, rather than just advising from the sidelines.

What Tasks to Delegate to the Agent

The main value of the update is that WordPress.com begins to function not as a closed admin panel, but as an environment you can have a conversation with. For teams with many repetitive actions, this is a noticeable time saver. Instead of manually creating a page, inserting blocks, moving text, and then separately managing discussion, an editor can describe the task in one series of commands and get a ready result for review.

• Creating draft posts from briefs or theses
• Assembling pages for product launches, mailings, or special projects
• Restructuring old materials and updating content blocks
• Moderating and managing comments from a single dialogue
• Preparing changes before publication with manual confirmation

This is especially convenient where the site operates in a fast cycle: news, company blogs, campaign landing pages, knowledge bases. You can give the agent not just a one-off command, but a chain of edits: shorten a headline, pull a list into a separate block, add a CTA, turn text into a page, check comments on a publication. The better the desired result is described, the more useful such a combination becomes. But final responsibility still remains with the editor or site owner.

Where Control Is Needed

Despite the word AI in the headline, the news is not about fully autonomous site management. And this is arguably a plus. Publication, editing, and moderation carry too high a cost of error: you can damage page structure, accidentally change text meaning, delete an important comment, or miss a reputational risk.

Therefore, the requirement to confirm each step looks not like a limitation, but a basic protection against unnecessary model overconfidence and failed automatic actions. There is also a more practical nuance: the quality of the result will depend not only on the model, but also on how carefully the site itself is organized. If sections, page templates, and content blocks are in disarray, the agent finds it harder to act predictably.

It might understand the task too broadly, restructure material more aggressively than needed, or propose a structure that is formally correct but does not align with editorial logic. Therefore, such a tool works best in teams that already have clear publication rules and basic CMS discipline.

What This Means

WordPress.com is taking another step toward making the CMS a working interface for AI agents, rather than just a data source. For editorial offices, marketing teams, and independent authors, this means a shorter path from idea to publication. Not instant autonomy, but managed acceleration of routine operations that can be built into the existing workflow without surrendering human control.