The internet is an iceberg.
I caught this mood at GoLive 2026, Pantheon’s inaugural conference, where Nick Thompson, CEO of The Atlantic, traded questions with Stephanie Cohen, Chief Strategy Officer at Cloudflare, Matthew Duncan, former Future of Work leader at Microsoft, and Jody Schwartz, SVP of People at Pantheon.
But Leo doesn’t have to die. We just have to pay attention to all the bot activity that is happening beneath the surface so we can navigate each of our Titanics appropriately.
Cohen from Cloudflare described a web where humans see the tip, while a much larger volume of activity happens underneath between agents. People may visit a handful of sites to answer a question. Agents can visit thousands, compare what they find, and decide what information reaches us next.
That image changes the marketer’s job. The question now goes beyond attracting a person to a page. It is how to make your organization understandable (and worth choosing!) when an agent is doing the first round of research.
The answer is not “make everything frictionless.” Matthew Duncan, former Future of Work leader at Microsoft, made that clear with a Walmart example.
The best experience is not always the shortest path
When ChatGPT introduced an instant-buy experience, Walmart tried it. The result did not perform as well as Walmart.com. Duncan said it generated about a third of the volume. Walmart then moved in a different direction, developing its own Sparky-style agent that connected the experience to customer logins and rewards cards.
That distinction matters. The first version optimized for removing steps. The second preserved the things that made Walmart useful and familiar, then made them easier to access through an agent.
As Duncan put it: “Better didn’t mean frictionless. It actually meant convenience, but in a way that I found value.”
That is a useful correction for web marketers. We can mistake fewer clicks for a better experience, even when those clicks carry memory, trust, preference, or proof of value. A rewards account is not just a login. It is context. A product comparison is not just a list. It is a decision shaped by priorities. A brand is not just a destination. It is a reason to believe one recommendation over another.
Everything has changed in the AI era of the web. Every company is at the crossroads. Take a listen to how Cloudflare’s been adopting.
Give agents something better than a product summary
An agent can summarize a page. It cannot invent the context that the page fails to provide.
That means web marketers should think beyond “Is this page optimized for AI?” and ask better questions:
- Does the agent understand who this product is for?
- Can it distinguish a core benefit from a feature list?
- Does it know what makes this offer different?
- Can it explain the tradeoffs, limitations, and conditions that matter to a buyer?
- Is there a clear next step for the person who receives the recommendation?
This is where distinctive content becomes operationally important. Original research, useful comparisons, customer evidence, clear product language, and a recognizable point of view give agents more to work with than generic category copy.
The goal is to create one strong source of truth that works for the machine-mediated journey and the human experience that follows it.
Treat the agent as a channel, not the destination
The iceberg metaphor also helps clarify what marketers should protect. The agent may become the discovery layer, but it does not have to become the entire brand experience.
An agent can introduce a buyer to your company. Your owned web experience still has to answer the next questions: Can I trust this organization? Does it understand my situation? Is the experience coherent? Can I get help? What happens after I click, sign in, or talk to someone?
This is why the Walmart example is more useful than a simple story about AI shopping. Walmart did not abandon its relationship with customers in the name of speed. It used an agent to extend a relationship it already understood.
For marketers, the work is to make that relationship legible at every layer:
- Build content that gives agents real context, not just keywords.
- Preserve the account, preference, and service signals that make an experience valuable.
- Create clear paths from an agent’s recommendation to an owned destination.
- Measure whether agent-mediated traffic produces useful outcomes, not just more requests.
- Keep updating the information agents rely on, because stale context creates stale recommendations.
At Pantheon, we spend a lot of time thinking about how people, machines, and web teams work from shared context. For marketers, that is the practical heart of the agentic web: your brand needs to be easy for agents to understand without becoming generic to the people who encounter it.
The web may be an iceberg. Your strategy should not be.
Check out where Pantheon’s headed in the AI era of the web.
Read our guide on how to architect an AI-ready website
