Think about the last website redesign brief you wrote, or received, or signed off on.
There were probably sections about colour palettes. Typography. Mobile responsiveness. Maybe a discussion about whether to use a page builder or go custom. Copy. Images. Some back-and-forth about the hero section.
What was not in that brief: permissions, actions, structured data feeds, approval gates, audit logs, rollback, or agent access controls.
It was not a gap. At the time, none of that mattered. Websites were for humans. Humans click, read, decide, and fill in forms.
That assumption is starting to break.
The next visitor to your website might not click anything at all.
Not because they are bouncing. Because they are an AI agent acting on behalf of someone: comparing your product against three others, checking availability, filling in a quote request, pulling your service specs into a buying decision, or booking a call without a human ever touching the keyboard.
This is not a fringe scenario anymore. The WebMCP draft Community Group report lays out the architecture for exactly this. Instead of forcing browser agents to navigate your website like an impatient intern clicking through menus, a site can expose structured tools: JavaScript functions with natural language descriptions and defined schemas that an agent can call directly. Book. Compare. Check. Submit. Update. Structured, permissioned, logged.
The developers have already noticed. Cloudflare has agent-facing documentation and MCP servers. Google is pushing A2UI and agent-ready interfaces. Lead-generation platforms are already building agent-facing capture surfaces alongside their existing web forms.
The business owners have not caught up yet.
Most of them are still arguing about AI content. Whether to use ChatGPT to write blog posts. Whether their new website will rank in AI search if they add a couple of FAQs. Whether the chatbot widget in the corner counts as AI-powered.
Those are fine questions. They are just not the important ones.
The important question is whether your website and your commercial system can be understood, trusted, and acted on by a machine.
That means:
- Your product information is accurate, complete, and structured. Not trapped in a PDF, scattered across a brochure, or dependent on a sales conversation to fill in the gaps.
- Your service pages tell the whole story: who it is for, what is included, what it costs or depends on, what proof exists, and what happens next.
- Your booking, quote, and contact flows can actually be completed without a human clicking around guessing what the next step is.
- You know who can do what to your website. Not probably the web team. Documented permissions, with clear production guardrails.
- You have logs. If an agent, or a human acting quickly, changes something, you can see it, review it, and reverse it.
I spent some time this week looking at two WordPress MCP tools, Novamira and Respira, and the contrast tells you everything about where this is going.
Novamira is powerful. Full PHP execution, file access, database, WP-CLI, the works. On a development environment, brilliant. On a client's live website, you are one mistaken agent action away from a very long afternoon.
Respira takes the opposite approach. It snapshots before editing, requires approval on significant changes, builds in rollback, and treats every agent action as something that needs a human in the loop for anything risky.
That is the only sensible framing: agent-ready does not mean give it the keys. It means designing clear, permissioned, recoverable access.
The same logic applies everywhere outside WordPress. Your product data becomes the truth layer that AI systems use to compare and recommend you. If it is inconsistent, different prices on different pages, product specs that contradict the feed, stock levels nobody updates, agents will get it wrong at scale. That is your problem, not theirs. Your support content becomes a knowledge surface. If it is three years out of date and buried in a PDF nobody updated, expect agents to confidently tell customers the wrong thing.
Here is the version of this that I think most businesses need to hear:
AI will not fix a messy business. It will make the mess faster and more expensive to untangle.
The companies that will benefit from the agentic web are the ones that get boring right first. Clear data. Honest offer pages. Permissioned systems. Proper documentation. A source of truth someone is actually responsible for.
That is not the flashy version of AI transformation. It will not make a great demo at a conference. But it is what the next website brief actually needs.
Not just: make us a better website.
Make our business legible and operable. For humans. For search engines. For AI systems. For the agents that are coming next.
That is the work. And most of it happens before you touch the homepage.