I am currently circulating a draft for WordPress project AI strategy vision. This post is a snapshot of that document. It has been born from mine & Grzegorz’s Knowledge merge proposal and it explains reasoning behind it.
The idea is to keep the open web alive in the AI world.
WordPress is the biggest open ecosystem to create your home on the web, configure it according to your needs, and host it exactly the way you like. The strategy we propose stays on the same message, but makes sure WordPress strengths shine in the AI world.
People are moving aspects of their daily work into Claude, ChatGPT and agent tools. Their connections, notes, skills, memories are piling up on platforms they don’t control. Interoperability, open protocols and secure access has to include agents and their needs.
WordPress should become the place where all your AI Agents and employees alike can store your private data and tools and where the knowledge work happens – in the place you control, extend and connect to external and internal AI providers.
“Content Management System” is only one letter away from “Context Management System”:
What this strategy unlocks
| With WordPress it’s easy, safe and extensible | Without WordPress it’s hard, messy and closed |
| User is able to manage their AI Skills database in WordPress, syncing their OpenClaw skills with Claude and ChatGPT. | Unmanaged directories of multiple sources of skill files with no UX |
| Businesses and enterprises can sync certain skills with their team, while other stay private | Distributing via enterprise features of Claude or asking the team to install specific repositories and asking them to learn git and sync directories. No UI to edit skills |
| User is able to safely expose certain tools (like for example calendar connection) to agents while making sure secrets stay safely managed by WordPress | Random packages installed from random skills on local machine, potentially leaking secrets or API keys. Collections of MCPs, potentially unsafe |
| User is able to switch from Codex, to Claude, to any other agent, ensuring all their stored knowledge, notes, skills, and configuration is ported | No good way – you can expose the same workspace to OpenClaw and Claude, but they have different file conventions and symlinks break (I tried). For syncing memory closest is something like GBrain |
| User is able to centralize all their MCP connections through WordPress, ensuring access policies follow centralized rules | Enterprise centralized MCP proxy services |
Plugin developers can create plugins that improve agentic memory, the experience of managing AI data (skills/memories/etc) or that experiment with novel AI uses. | Because each plugin owns its data, monolithic plugins are incompatible with each other. |
| WordPress-native agents are immediately as capable as Claude / Codex due to shared knowledge and content already available on your website | You have to teach each agent from scratch |
Why WordPress can do this
- Memories, Documents, Skills, and all Markdown files are all content. WordPress is from ground up a Content Management System, with appropriate data types, screens our users are familiar with.
- It has a robust and tested permission system, with your team already using it. Data provenance, access control and user education are a solved problem
- It already has an ecosystem of business capabilities and plugin marketplace much richer than Claude / ChatGPT ones. We can leverage existing connections and APIs to gain network effects from plugins.
- WooCommerce is a good example: It stores financial data, runs orders and exposes mission-critical data flows.
Future-proofing WordPress
We are seeing a new wave of software: Claude Cowork, Codex for office work, Office 365 Copilot, Google Workspace Gemini, OpenClaw. All these aim to integrate agentic capabilities with your data, skills and AI memories to provide better AI help while also hosting your data on their platform. This creates 2 risks:
- As these companies compete they may decide to lock the private data
- Eventually, as these products consume more work tasks, these companies will offer website hosting, as we see with ChatGPT Sites.
WordPress stays relevant in that future if we:
- Establish a vision that will allow our experiments to share the same broad direction and allow independent explorations to stack upon each other.
- Unlock new use cases for WordPress, as AI is changing user behavior. While WordPress has a lot of advantages, users who do their work inside agent platforms have less reason to visit their WordPress site.
- Commit to not forcing AI. This document assumes no changes in the classic WordPress experience. The cornerstone of this strategy is to make subtle, behind-the-scenes changes that unlock new possibilities, but require no change in habits.
That’s a good idea. Yeah. Sounds like a good product
– Stephen Wolfram, Automattic advisor when presented with this plan. This feedback made me really happy.
Workstreams
None of these workstreams are new. This document only describes how they fit together and how their development should fit with the overall needs of the vision.
WP AI Client
WP AI Client provides integration points for AI features running directly inside WordPress.
- ✅Plugin developers can focus on user-facing features or tweaks to the functionality instead of repeating boilerplates.
- ✅Users can choose any token provider they want
Abilities API
✅Abilities API is the equivalent to AI tools, with built-in permission schemes. Plugin developers can create WordPress-native equivalents to CLI tools for agents or custom MCPs, with credentials safely managed on WordPress.
🆕Abilities Catalog needs to solve ability orchestration problem by automatically organizing independent abilities
WP Knowledge
🚧 WP Knowledge (Merge proposal) aims at providing a common standard to store skills, memories, and notes on WordPress. It seeks to become a standard in Core so that plugins standardize on the same data format and are interoperable between each other.
WP AI Client introduced a way to call LLMs into WordPress, and Abilities API introduced tools to interact with WP itself. But developers wanting to build compelling AI features need to store memory, skills or work-in-progress artifacts somewhere, requiring external services or limiting capabilities of their plugins. Knowledge introduces a missing component: state. This completes the first phase of Core AI rollout, giving WordPress-based agents APIs to compete with other agentic solutions.
Finally, WP Knowledge is part of the AI Strategy, but it unlocks use cases for non-AI users too. A common API for knowledge bases and private data allows unrelated plugins to introduce features known from apps like Notion, Evernote, Google Docs and Automattic’s P2 system. This post describes these new opportunities extensively.
This document has been born to explain the reasoning behind WP Knowledge and how it fits with the broader context.
Interoperability
The core of this strategy is that you can access your tools, memories, notes, artifacts, configurations through any agent on any surface. This requires ensuring WordPress supports broadest array of standards and access points possible.
- ✅ MCP Adapter will allow triggering any ability through MCP-enabled clients
- 🆕 We need to expose WP Knowledge through MCP as well
- 🆕 We need to look into CLI thin client / sync tools for skills, memories and agents.md configurations
- 🚧 There are explorations of GIT protocol as sync layer.
Data Provenance
While WordPress will enable access to any data and tool for any agent you want, it also needs to ensure that private data stays private.
- ✅ Capabilities API manages access control for Abilities and Knowledge.
- 🆕 We need to explore finer access controls for MCP and Abilities instead of inheriting your (potentially admin user) permissions. This would turn WordPress into enterprise-grade permission proxy for all AI use. Some approaches include:
- Delegated users
- Access control matrix for specific agents
Vector Search
🚧 Vector Search API in WordPress would allow multiple plugins and solutions to search common infrastructure. RAG solutions are a very common request in the community and would strengthen AI, SEO and private search/research capabilities of WordPress.
This also combines and unlocks superpowers for Knowledge – like semantic skill search.
A safe way to proceed with this work would be to introduce a RAG connector that bootstraps an appropriate canonical plugin on the connectors’ screen.