Skip to main content
Trends & Strategy6 min read

Your Website May Soon Need a Version for Agents

August 26, 2026By Ajan Kanagalingam

OpenAI is pushing a standard that would let websites talk directly to AI agents, declaring what actions they offer rather than making an agent read the page and click things like a person. It is early, it is a proposal with a developer challenge attached, and it may not survive in this form. It is still worth noticing, because websites have been built on one assumption for thirty years, which is that a human is looking at them.

Agents are pretending to be people

Right now, when an agent uses your website, it reads the page and clicks things the way a person would. That fails in predictable ways. It misreads layouts. It gets lost in unusual navigation. It breaks the week after you redesign.

It is not much better from your side. You get traffic you cannot identify, forms filled in by something that is not a customer, and no say in how you are represented to whoever sent the agent. Nobody chose this arrangement. It is just what happens when software built for one kind of visitor meets another kind.

What a declared interface changes

The proposal is that a site states what it can do. Booking an appointment is an available action, here is what it needs. Checking availability is another. The agent stops guessing and the business decides what is on offer.

TodayWith a declared interface
Agent guesses which button booksSite says booking is an action
Breaks when you redesignSurvives a redesign
You cannot tell agents from peopleYou choose what agents may do

That third row is the one business owners should care about most, because it converts an uncontrolled situation into a decision.

You do not need to build anything

Let me be direct about this, because someone will try to sell you a project on the back of this news. It is a proposed standard. Adoption is unproven. Building for it now means building for something that may change shape entirely, and if it does become real, your website platform will almost certainly support it before you would have got round to implementing it yourself.

The genuinely useful step is cheaper and pays off either way: make sure the information on your site is accurate, current, and structured sensibly, because that is what assistants use to describe you today. We set out the approach in optimising for AI search, and it is worth doing whether or not any of this arrives.

Who should watch more closely

If customers complete something on your website, this matters to you sooner. Booking an appointment. Ordering. Requesting a quote. Checking whether you have something in stock.

Those are precisely the actions an agent would want to perform for someone, which means being agent-accessible could translate into actual bookings rather than remaining a technical curiosity. It is the same direction as assistants becoming a place where transactions happen, covered in customers ordering inside AI assistants. If your site is a brochure that leads to a phone call, this is a story to follow rather than a project to start.

Decide what an agent may do

If this does arrive, there is one decision worth having thought about in advance. There is a real gap between letting an agent check your availability and letting it make a booking on someone's behalf.

Reading is low risk. Acting is not, and the failure modes are things like a booking made by an agent for a customer who did not quite mean it. That is the same read-versus-write distinction we drew in capability shipping ahead of controls, arriving now on your own website rather than inside your systems. Default to read-only when the option appears, and add actions deliberately.

The thirty-year assumption

Everything about a website assumes eyes. Layout, visual hierarchy, colour, the position of the button. All of it exists to guide a human through a decision, and we have spent three decades getting good at it.

An agent needs none of that and is actively hindered by some of it. Whether the answer turns out to be WebMCP or something else entirely, the pressure to serve two very different kinds of visitor is not going away. That is worth understanding now even though it requires nothing of you today, because the businesses that get caught out by these shifts are usually the ones who first heard about them when a competitor had already adapted.

Frequently Asked Questions

What is WebMCP?

OpenAI is promoting a way for websites to expose their functions directly to AI agents, rather than making an agent read the page and click things like a person would. Instead of an agent guessing which button books an appointment, the site would declare that booking is an available action and describe what it needs. Treat the specifics as early, since this is a proposed standard with a developer challenge attached rather than something already widespread. The direction is what matters.

Why would a website need this?

Because agents currently interact with sites by pretending to be people, which works badly in both directions. The agent misreads layouts, fails on unusual navigation, and breaks whenever you redesign. The business gets traffic it cannot identify, forms filled by something that is not a customer, and no control over how it is represented. A declared interface fixes both problems: the agent knows what actions exist, and the business decides which ones to offer.

Does my small business need to do anything now?

Almost certainly not. This is a proposed standard, adoption is unproven, and building for it today means building for something that may change shape entirely. What is worth doing now is much cheaper and useful regardless: make sure the information on your site is accurate and structured, since that is what assistants use to describe you today. If a standard like this becomes real, your platform will most likely support it before you would have implemented it yourself.

Who should pay attention sooner?

Anyone whose customers complete a transaction on the website: booking an appointment, ordering, requesting a quote, checking availability. Those are exactly the actions an agent would want to perform on a customer’s behalf, and they are where being agent-accessible could translate into real business rather than a technical curiosity. If your site is a brochure that leads to a phone call, this is a story to watch rather than a project to start.

What is the risk?

Two worth naming. If agents can act on your site, you need to decide what they may do, because the gap between letting something check availability and letting it make a booking is significant. And you become dependent on how the assistant chooses to present you, which is the same landlord problem as any platform. Both are manageable, and both are reasons to approach this deliberately rather than enabling whatever your platform switches on by default.

Be found and described accurately

We help Canadian businesses structure their web presence so AI assistants represent them correctly, and decide what agents should be allowed to do.

Related Articles

Trends & Strategy

Why the Smart Money Is on Vertical AI Agents

August 3, 2026Read more →
Trends & Strategy

AI Agents and Your Next Software Renewal

July 30, 2026Read more →
Trends & Strategy

AI Agents Are Getting Logins and Wallets: Why It Matters

July 8, 2026Read more →
AK
Ajan Kanagalingam
Founder & ChatGPT Consultant, ChatGPT.ca

Ajan leads the ChatGPT.ca team: 200+ custom GPT builds and automation projects for 50+ businesses across 20+ industries. Based in Markham, Ontario. PIPEDA-compliant solutions.

Stay ahead of AI in Canada

Weekly case studies, new tools, and ROI playbooks for Canadian SMEs. One email, zero spam.