Microsoft said the quiet part out loud this month.
Not with a think piece. Not with a vague keynote line. With product architecture.
On June 2, Microsoft introduced Web IQ and described it as a web search stack built for AI systems and agents, not just for people typing queries into a search box. That matters because it turns a trend into a declaration. One of the biggest companies on the internet is no longer treating search as a single thing. It is treating search for humans and search for software as two different jobs.
Google is making the same bet from the other direction. On May 19, Google said AI Mode had already passed one billion monthly active users globally. It also said the average AI Mode query is three times the length of a traditional search query, and that more than one in six searches in the US now use voice or images. Then Google added another important detail: starting in late June 2026, Android devices will get Gemini in Chrome and Chrome auto browse.
That means the biggest browser ecosystem on earth is moving closer to delegated action, not just better retrieval.
A page is for people. Evidence is for software. The new web needs both.
This is why I think a lot of the current AI-search conversation is still too small. People keep arguing about whether AI Overviews are good, whether SEO is dead, or whether agents are overhyped. Those are surface-level questions. The deeper thing happening is simpler: the web is splitting into a human-facing interface and an agent-facing interface.
And once you see that, a lot of recent product moves stop looking random.
Microsoft just named the split
The most revealing thing about Web IQ is not that Microsoft built another search API. It is how Microsoft framed the problem.
Traditional web search is designed for a person who asks a question, gets results, clicks around, compares pages, and decides what to do. Agent search is different. Agents retrieve over and over, compare evidence, update plans, and stay inside a latency budget because every extra step costs time and tokens. That is not a minor tweak to search. That is a different workload.
Once the workload changes, the output changes too. Humans can handle pages, menus, sidebars, and a little mess. Agents want clean evidence, narrow passages, reliable attribution, and action surfaces that are easy to use without improvising around broken UX.
So when Microsoft talks about passage-level evidence, low-latency grounding, and optimizing for repeated retrieval inside reasoning loops, it is effectively admitting something huge: the pageview is no longer the unit of value in the same way it used to be.
For decades, the default internet loop was crawl, rank, click, visit, monetize. The new loop is starting to look more like retrieve, ground, decide, act, and only sometimes visit.
That is not a side experiment anymore. That is a distribution event.
Google is normalizing delegated browsing
Google's AI Mode numbers matter for one reason above everything else: they show normal users are already behaving more like managers of software than classic searchers.
Longer queries are not just a stat. They are a behavior shift. When users move from typing fragmented keywords to asking planning questions, comparison questions, and multi-step questions, they are treating the system less like a library index and more like an operator they can brief.
That gets even more obvious when you add the Chrome roadmap on top. Gemini in Chrome helps summarize, compare, and research. Chrome auto browse takes on repetitive tasks directly. That means browsing is not just becoming conversational. It is becoming delegable.
I wrote earlier this year in Google Wants Search to Be Your Operating System that Google was trying to move from blue links to intent infrastructure. That thesis looks stronger now, not weaker. Search is no longer just about helping you find a page. It is about helping the software around you understand what you want, gather what it needs, and increasingly do the task for you.
That changes the role of the browser. The browser used to be a window. Now it is starting to look like a junior operator.
The old web bargain is starting to crack
The open web worked on a simple bargain for a long time. Search engines could crawl your content because they sent people back to you. Publishers tolerated the extraction because they still got audience, ad impressions, subscriptions, leads, and brand lift on their own turf.
The agent web puts pressure on that bargain from both sides.
On the retrieval side, agents want the best parts of a page without the ceremony around the page. On the action side, they want to complete the task without making a human bounce through five interfaces and three popups. From the user's perspective, that is better. From the publisher's perspective, it can mean more extraction and less direct relationship.
That is why the fight around AI search keeps escalating. It is not just about citation style. It is about economics and leverage. I already hit one version of that in The UK Just Forced Google to Admit AI Search Needs Publishers. The moment regulators started forcing opt-outs, metrics, and clearer attribution, the market signal became obvious: even the platforms know this new model cannot run forever while starving the sources it feeds on.
Cloudflare sees it too. During Agents Week, it talked openly about agents driving a growing share of internet traffic and about websites needing new tools to become agent-ready. That is not fringe theory. That is infrastructure reacting to traffic reality.
If the human web was built around sessions, pages, and persuasion, the agent web is being built around state, evidence, and executable actions.
Most websites still only speak human
This is where a lot of companies are going to get caught flat-footed.
Most websites are still optimized for a human who can tolerate ambiguity. A person can infer missing context. A person can scroll around bad layouts. A person can mentally connect the hours page, the pricing page, and the booking form even if they all contradict each other a little.
Agents are much less forgiving.
If your key facts only appear inside a hero paragraph, a lazy-loaded tab, and a JavaScript-heavy widget that breaks on mobile, you do not have a robust information surface. You have a vibes surface.
If your policies are inconsistent, your canonicals are sloppy, your structured data is missing, your contact details drift between pages, and your product details are buried in marketing copy, human visitors might still muddle through. Agent systems will either miss the facts, mistrust the facts, or skip you for a cleaner source.
That is why I think a lot of "GEO tips" are missing the point. This is not about stuffing your site with AI bait. It is about making your site legible as a system.
- Clear facts that do not contradict each other
- Stable URLs and canonical signals
- Structured data where it actually helps
- Explicit action paths for booking, buying, contacting, or subscribing
- Publisher controls and machine-readable policies that are not an afterthought
Humans browse. Agents reconcile. If your site cannot survive reconciliation, the next generation of discovery will be rough.
The winners will serve both audiences at once
I do not think the human web disappears. People still want design, trust, voice, aesthetics, and the feeling of choosing. A webpage is not just a data packet. It is also brand, tone, proof, and experience.
But the websites that win from here are going to do two jobs at once.
They will feel good for humans and read cleanly for software. They will have strong editorial voice and strong machine legibility. They will expose enough structure that an agent can ground itself correctly, but enough style that a person still wants to land, stay, and care.
That is also why I do not buy the lazy take that AI makes voice irrelevant. It probably makes blandness irrelevant faster. Commodity pages get compressed into evidence. Distinct viewpoints still travel.
If everyone can extract facts, then what remains scarce is judgment. What remains memorable is perspective. What remains durable is trust.
The human web still needs people worth reading. The agent web just changes how those people get found and routed.
Stop thinking in pageviews. Start thinking in evidence
The big mental shift here is that a lot of site owners still think the goal is to win the click. In the next phase of the web, the more useful question is whether your site can produce trustworthy evidence and clean action paths inside somebody else's reasoning loop.
That sounds less glamorous than ranking No. 1. It is also closer to reality.
If an agent compares five providers, can it understand your offer without hallucinating? If a browser-side assistant tries to book an appointment, can it actually complete the flow? If an AI search product summarizes your expertise, is there enough structure and clarity for it to quote the right thing for the right reason?
That is the new interface layer. And it is forming in public right now.
So no, I do not think the web is dying. I think it is bifurcating. One layer is still about human attention. The other is about machine consumption and delegated action. Companies that understand both will adapt. Companies that cling to the old crawl-click-pageview religion as if nothing changed are going to mistake traffic erosion for bad luck.
There is now one web for people and another for agents. The smart move is to build for both before one of them quietly routes around you.