Every executive at every major company in the world has the same goal this year: use AI. Use it in the product, use it in operations, use it in everything. The mandate is universal; the path to executing it is not.
For digital teams, the pressure can land in many places. The marketing organization wants AI-assisted content production and more personalized experiences. The IT organization wants AI tools that operate cleanly inside existing governance — without one-off pilots, sprawling vendor contracts, or surprise data exposure. Both sides want what the CFO wants: a way to act on the mandate without an open-ended bill.
WordPress 6.9 and 7.0, released in December 2025 and May 2026, were built for this moment. Together they introduce a coherent set of AI Building Blocks — the Abilities API, the AI Client, and the MCP Adapter — that turn an enterprise CMS into a place where AI capabilities can be added, governed, and swapped without bespoke engineering for every model and every use case. For organizations already running WordPress at scale, this is the difference between watching the AI conversation and being part of it.
What Changed in WordPress 6.9 and 7.0
WordPress now has a shared foundation for leveraging AI on any site. Three new pieces — an action layer, a model layer, and an agent layer — are first-class parts of the platform or official adjacent packages. Your site can build on top of those foundational layers instead of reinventing them.
The Abilities API: Making the Site Itself Addressable
The Abilities API shipped in WordPress 6.9 in December 2025, and was extended further in 7.0. It gives WordPress a standardized way to describe to an AI model what your site can actually do — publish a post, schedule content, update a product, or moderate a comment.
Each Ability has clearly defined inputs, outputs, and permissions. AI tools and automation systems can ask the site what it can do, and the site can answer in a format those tools understand. Out of the box, WordPress 6.9 ships with three read-only core Abilities — site info, current user info, and environment info — as a starting point that you can extend with plugins or your own custom code.
What that means for an enterprise buyer:
- Governance is built in. Each Ability declares who can run it, what data goes in, and what data comes back. An AI assistant that can summarize comments does not automatically inherit the right to delete them.
- A single source of truth for what the platform can do. New plugins extend the catalog; existing plugins start to interoperate without custom glue code.
- Future-proofing for the agent era. Capabilities defined this way work whether the consumer is a plugin, an internal automation script, or an external AI agent.
The AI Client: One Connection Point for Any Model
The AI Client is a built-in interface in WordPress 7.0 that lets any plugin or your own code send prompts to any supported model without writing model-specific code. Switching from Anthropic to OpenAI to Google is a configuration choice, not a development project.
Connections are managed in a new Connectors screen inside the CMS. WordPress core deliberately ships without any AI provider bundled — instead, three official free provider plugins are maintained by the WordPress project (for Anthropic, Google, and OpenAI), and the same pattern is open to any other provider. The reason: AI providers change too fast to be baked into a semi-annual core release cycle.
What that means for an enterprise buyer:
- No model lock-in. If pricing, performance, or compliance posture shifts on one provider, you change a setting. You do not replatform.
- One place for credentials. API keys live in a single, auditable surface inside wp-admin instead of being scattered across half a dozen plugins.
- A common foundation for every AI feature on the site. Content tools, search tools, moderation tools, and editorial workflows all draw from the same underlying connection.
The MCP Adapter: Making AI Assistants Talk to Your CMS
The WordPress MCP Adapter — an official WordPress AI team package — turns the Abilities your site exposes into tools consumable by any AI assistant that speaks Model Context Protocol. Today that includes Claude Desktop, Claude Code, Cursor, and VS Code, with a growing list of others. WordPress can act as an MCP server (your site as a set of tools an AI can call) or, in the right configuration, as an MCP client (your site reaching out to other AI-aware services).
Once a site speaks MCP, an editorial team can ask their AI assistant to draft a story, pull related coverage, and queue it for review — and the assistant can do all of that against the actual CMS, with the actual permission model in place. The MCP Adapter’s own guidance recommends starting conservatively: read-only abilities first, dedicated MCP user accounts, and tight permission checks. WordPress remains the system of record; the AI is a privileged client, not an unsupervised one.
What that means for an enterprise buyer:
- Open standards rather than a vendor’s proprietary agent layer. MCP is being adopted broadly across the AI ecosystem; building on it hedges against any one AI company’s roadmap.
- One implementation, many assistants. Your team can use whichever AI tool fits their workflow. The site does not care.
- A measured path to agentic workflows. The same Abilities your site already exposes for plugins are now available to external AI assistants — under your governance, your auth, and your audit trail.
One operational caveat worth raising at the buyer level: today, self-hosted sites typically authenticate AI clients via WordPress application passwords, with OAuth 2.1 already live on WordPress.com and in development for self-hosted environments. Enterprises with strict identity and access requirements should plan around that trajectory rather than against it.
What This Unlocks in Practice
The clearest near-term use cases for enterprise teams already running WordPress:
- Editorial assistance inside the editor. Draft, summarize, suggest headlines, generate alt text, and translate against the model you choose.
- Smarter audience-facing experiences. Conversational search, related-content recommendations, and personalized recirculation, all sharing one model connection.
- AI-aware operations. Connect WordPress to your existing AI assistants for content audits, SEO sweeps, compliance reviews, and editorial reporting — without exporting data to a third-party tool.
A Note on What’s Still Your Job
WordPress 6.9 and 7.0 make AI infrastructure standard. They do not make AI strategy standard. You still need to decide where AI belongs in your workflow, which use cases pay back the cost of the model calls, and how to evaluate the output. The new infrastructure shortens the integration runway from months to days. The thinking is still on you.
Key Takeaways
- The “use AI” mandate now has matching infrastructure in WordPress. The Abilities API (6.9) plus the AI Client and Connectors API (7.0), together with the official MCP Adapter package, create a coherent AI foundation.
- Model choice is a configuration decision, not an engineering project. Anthropic, Google, and OpenAI ship as free official provider plugins; others can be added.
- Governance, permissions, and discoverability are built into the platform. AI assistants can only do what your site explicitly exposes them to do.
- WordPress is a credible host for agentic workflows. Through MCP, the same capabilities are available to any AI assistant your team uses — under your authentication and audit controls.
The technical lift is dramatically smaller than it was twelve months ago. Enterprises already running WordPress at scale have the shortest path to a defensible answer when leadership asks, “what are we doing about AI?”
