A multi-agent AI workflow is a few model calls in a row, each with its own role prompt. The pattern itself is simple. What decides whether it survives on a real WordPress site is the queue behind it, and what each agent actually sees.
Automattic shipped ten experiments during Radical Speed Month, and most of them only exist on WordPress.com. The two that don’t are Desktop Mode and Studio’s Blueprints Gallery. Here’s how to sort the list, what a blueprint file looks like, and what I’d actually adopt.
Nathan Bos argues we should train AI to betray its users, because obedient superintelligence is a lone bad actor’s dream tool. I agree with where he ends up and disagree about the layer: for sites and stores, that decision belongs in permissions code, with a human on the send button.
An AI assistant that can’t see your project turns every review into a copy-paste loop. What I’d build myself, what I’d leave to the official SDK, and where the sandbox boundary belongs when the root contains wp-config.php.
A hundred lines of Python, no vector database, and a real PDF. That is the baseline this author ships, with answers that cite exact lines. The part worth borrowing for WordPress work is the answer schema: typed JSON, verifiable quotes, and a null path that refuses to invent.
WooCommerce is shipping read-only abilities across Subscriptions, WooPayments, shipping and marketing extensions, and it published the agent skills used to build them. I’d skip the hype and use those skills on your own plugin: install, audit, register behind a version guard, then verify.
Agents are useful on WordPress work until the day one runs wp plugin update –all on a live store or rewrites the database without a dump. This is the WordPress version of the blocked-commands list: what to gate, the WP-CLI flags that help, and the AGENTS.md rules I write for client repos.
A founder gave ChatGPT an intern’s month-long data problem and got 80 percent of a prototype in 48 hours. I look at what that means for WordPress research spikes: the demo you get, the shortcuts agents take, and the understanding you quietly skip.
Combining two coding agents works when the second one reviews what the first wrote. I explain how I split Claude Code and Codex across WordPress client work, which repo rules the reviewer needs, and where staging still beats both.
Clients keep asking for AI product search on stores where the catalog is already a database. Before anyone writes an embedding job, run the plain query and see what it answers. Here is where RAG genuinely fits on a store, and where a filter wins.