Have You Noticed? The Internet is Changing
Not in the slow, cumulative way it usually does. In a "the numbers just crossed a line we didn't expect to cross yet" way.
In June 2026, Cloudflare's CEO said something that should have made bigger news than it did: automated traffic had overtaken human traffic on the web for the first time in its history — roughly 57.5% bot, 42.5% human. Cloudflare hadn't expected that crossover until 2027. It arrived early, and agentic AI is the reason.
Sit with that for a second. Most of what's hitting your website right now isn't a person.
The web is getting a new front door, and it isn't your homepage
For twenty-five years, the customer journey has run through pages built for human eyes: a homepage, a nav bar, a search box, a form. That's the layer marketers have spent our careers optimizing.
A second layer is now forming on top of it. AI will research, compare, and buy a product without the user ever landing on the merchant's site — Instant Checkout and the Agentic Commerce Protocol already route purchases through Etsy and a growing list of Shopify brands this way. Salesforce is pushing the same logic into the B2B side with Agentforce and, as of this year's TDX conference, "Headless 360" — decoupling the interface from the system so an agent can act on the backend directly. Google, Perplexity, and Microsoft are all shipping browsing agents that click, fill, and transact on a person's behalf.
The lead isn't always a human scrolling your site anymore. Increasingly, it's an agent sent to gather answers, and it's making judgment calls about your business before a person ever sees your brand.
This layer is going to get bigger, not smaller
Whatever version of this you're seeing today is the early one. A few things are converging to accelerate it:
Standards are showing up faster than most teams can track — the Model Context Protocol (MCP) for connecting agents to tools and data, the Agentic Commerce Protocol for transactions, llms.txt for telling agents what a site actually is. Every major platform vendor is racing to be the rail these agents run on, which is exactly what happened with mobile and cloud before this.
The buyer journey is compressing. Research, comparison, and purchase — steps that used to happen across multiple sessions and multiple humans on a buying committee — are collapsing into a single agent-mediated exchange. That changes what "the funnel" even means.
And the agents aren't just consumer-facing. Enterprise buying committees are starting to deploy their own agents to evaluate vendors, meaning your next RFP might be partially read by software before a person opens the deck.
Headless is the quiet infrastructure story underneath all of it
None of this works without a shift most customers will never see: systems talking to systems, with no page in between.
Headless architecture — content, commerce, and data exposed through APIs rather than locked inside a rendered webpage — used to be a developer preference for flexibility and speed. It's now the precondition for agent access. An AI agent can't "browse" a monolithic site the way a human does; it needs structured, machine-readable access to the same content and transactions. That's why every major platform announcement this year — CMS, CRM, commerce — has led with headless and agent-facing APIs rather than UI.
The practical result: more of the internet's actual work will happen server-to-server, agent-to-agent, with no human-facing screen involved at all. The website stops being the only interface to your business and becomes one interface among several.
What this means for the rest of us
If you work in marketing, revenue, or GTM, this isn't a future-roadmap item — it's already live in your analytics, whether you've labeled it correctly or not.
A few questions worth asking this quarter: Is your content and product data structured well enough for an agent to accurately represent you, not just for a human to skim? Do you have any visibility into agent traffic versus human traffic, or is it all bucketed as "other"? If a buying committee's agent is the first "visitor" to evaluate you, what does it find — and is it accurate?
The internet spent three decades being built for human attention. The next chapter is being built for machine interpretation, with humans approving the final decision. That's a different discipline, and most of us are still optimizing for the old one.
Have you noticed it yet in your own numbers? I'd like to hear what you're seeing.
*Views are my own and don't reflect the positions of my current or past employers.
