The Agentic Web

The agentic web is here. Your next customer might send a bot first.

In mid-2026 Google, ChatGPT, and Claude all shipped AI agents that browse websites, compare options, and even book appointments for a person. Your site now has two audiences: humans, and the agents people send ahead of them. Here is what changed, and the honest, no-hype way to get ready.

By Adi Huric, founder of Most AI Labs·July 2026·8 min read

For twenty-five years, a website had one job: be readable by a person. In 2026 that quietly stopped being true. Your site now gets two kinds of visitors. The first is a human. The second is an AI agent the human sent ahead to do the looking, and increasingly, the booking. The agent arrives first. It does not scroll, squint, or give you a second chance.

This is not a prediction. Between May and July 2026, the three biggest AI players all shipped agents that visit real websites and act on them. Here is exactly what landed, why it matters for a Vancouver business, and the part almost every agency is getting wrong right now.

What actually shipped in mid-2026

Google, at I/O 2026. Google called it “the biggest upgrade to our Search box in over 25 years.” AI Mode now runs on Gemini 3.5 Flash by default. More importantly, Google launched information agents that run in the background and monitor the web for you, and it began expanding agentic booking for local services like home repair, beauty, and pet care, including Google calling businesses on the user’s behalf to make the appointment. AI Overviews are now reported to reach over 2.5 billion people a month across roughly 200 countries.

Anthropic’s Claude. Claude Sonnet 5 shipped on June 30, 2026 and became the default for every free and paid user the next day. It can drive a browser and a terminal and run on its own long enough to complete real tasks, the kind of work that a few months earlier needed a far bigger, slower model. Claude’s paying user base has grown roughly 75% since January 2026.

The pattern underneath both. Across the industry, people are shifting from chatting with AI to delegating to it. You no longer ask the assistant a question and then go do the task. You hand it the task. That agent is now the thing that reads your website.

Key takeaway

Key takeaway: the first visitor to your site is no longer always a person deciding whether to call you. It is often an agent deciding whether to include you in the shortlist it hands back to a person, or the two or three options it offers to book directly.

Why this hits local businesses first

A person searching “dog groomer near me” used to open five tabs and skim. Their agent does not open tabs. It reads the underlying page, pulls out your services, hours, service area, and price signals, checks your reviews, and forms an opinion in about a second. If it cannot find those facts in plain, machine-readable text, it does not email you to ask. It moves to the competitor it could read.

And when Google’s own system offers to phone a business to book a grooming or repair slot, the businesses it can call are the ones whose booking and contact details it could parse cleanly in the first place. The bar is no longer “does my site look nice.” It is “can a machine understand what I sell, who I serve, and how to act.”

One honest caveat, because we do not do hype: for genuinely local intent, AI answers are still the minority of results. Industry tracking puts AI Overviews on roughly 7% of local searches today. The Google Business Profile and Map Pack still carry most local visibility. The agentic shift is real and accelerating, but it is a growing slice, not the whole pie yet. Plan for both.

The part almost everyone is getting wrong

Right now a wave of agencies is selling a panic product: a special file called llms.txt, plus “AI schema” and “AI-specific rewrites,” pitched as the secret key to the agentic web. It is a tidy story. It is also, according to Google itself, not how this works.

In its 2026 guidance on AI features, Google explicitly mythbusted the idea. In its own words, “optimizing for generative AI search is optimizing for the search experience, and thus still SEO.” Google states plainly that you do not need to create llms.txt files or special markup for AI, that structured data is “not required for generative AI search and there’s no special schema,” that you do not need to chop your content into AI-sized chunks, and that chasing inauthentic brand mentions “isn’t as helpful as it might seem.”

Watch out for this

If someone quotes you a fee to “optimize your site for AI” and the deliverable is mostly an llms.txt file and some AI schema, you are paying for a myth. The file is cheap and harmless to add, but it is not the lever, and Google says so directly.

The agentic web did not invent a new rulebook. It raised the stakes on the old one. Everything that made your site legible to Google now decides whether a machine will act on it.

What actually gets you ready

The real work is unglamorous and durable. It is the same foundation that helps human visitors and Google ranking, which is exactly why it is worth doing. In priority order:

·Put the facts in real text, not in pictures or scripts. Your services, hours, service area, and contact details must be actual HTML text an agent can read, not baked into an image or loaded only after heavy JavaScript. This is the single most common failure on small-business sites.
·Be fast and lightweight. AI systems deprioritize slow, bloated pages because they “cost more to process and risk timing out before retrieval even finishes.” The 2026 targets: Largest Contentful Paint under 2.5 seconds, Interaction to Next Paint under 200 milliseconds, Cumulative Layout Shift under 0.1. Server-rendered HTML beats a heavy single-page app here.
·Write non-commodity content. Google’s guidance rewards pages with unique insight over generic copy anyone could have written. A real answer to a real customer question outranks a page of keywords.
·Keep your Google Business Profile and any product feeds accurate. For local and commerce, these structured sources are what the agents actually pull from. Google names Business Profiles and Merchant Center feeds directly.
·Make action easy to complete. A clear, standard booking flow and a plainly labeled phone and address let an agent (or Google calling on a customer’s behalf) finish the job instead of giving up.
·Earn real reviews and real third-party mentions. Trust signals from places Google already trusts are what tip a machine from “maybe” to “recommend.” This is covered in depth in what an entity is, and how the internet measures trust.

Notice what is not on that list: no secret file, no special schema, no rewrite-for-robots. If you do want to add an llms.txt file, go ahead, it costs nothing and does no harm. Just do not mistake it for the work, and do not pay a premium for it.

The bottom line

·In mid-2026, Google, Claude, and ChatGPT all shipped agents that browse and act on real websites. The agent now reads your site before the human does.
·For local businesses this is a growing slice of first contact, not yet the majority. Prepare for it without abandoning your Google Business Profile and Map Pack.
·Ignore the llms.txt panic. Google itself says AI search is still SEO, and that special files and schema are not required.
·The prep is the durable fundamentals: readable text, fast pages, genuine content, an accurate Business Profile, easy actions, and real trust signals.

The fastest way to find out whether the agents can already see you is to ask them. Our free AI Visibility Score runs the same query across ChatGPT, Claude, Gemini, and Perplexity in about 60 seconds and shows whether your business appears, who it appears next to, and what each engine says about you. If you want the full picture, the free 7-day audit tests your site the way an agent would: what a machine can read, how fast it loads, and where the gaps are against your top competitors.