The internet was built for people. Now it must bend for the machines that never sleep.
AWS quietly launched the next generation of its OpenSearch Serverless service on May 28, 2026. The update decouples compute from storage so the system can spin up in seconds to handle sudden bursts from AI agents and then drop to zero when those agents vanish. No more paying for idle capacity that human users rarely demanded. Amazon Web Services designed the change explicitly for workloads that spike without warning and idle without notice.
Tia White, general manager for Amazon OpenSearch Service, put it plainly. “The timing is straightforward. Agents are moving from experimentation into production, and they create traffic patterns that previous infrastructure simply wasn’t designed for. They spike without warning, they go idle without notice, and enterprise needs search that keeps up without paying for empty or idle compute.”
Short. Direct. And already happening at scale.
Cloud infrastructure grew up around predictable human behavior. Click. Scroll. Stream. Pause. AI agents do none of that. They spawn sub-agents. They hammer databases. They chain API calls at machine speed before disappearing. The result is traffic that looks nothing like the steady flow cloud providers engineered for over two decades.
Data tells the story. Bots already accounted for 31 percent of overall HTTP traffic in the six months leading into 2026, according to Cloudflare. AI crawlers, search engines and assistants made up roughly one quarter of those bot requests. “Non-human traffic will exceed human traffic sometime in the first half of 2027,” Lai Yi Ohlsen, senior product manager at Cloudflare, told TechCrunch.
But the change runs deeper than raw volume. Google used its I/O conference days earlier to preview how users will hand off research, booking and browsing tasks to AI systems. Enterprises now deploy agents internally to query data, summarize reports and trigger actions across tools. Each deployment multiplies machine-to-machine conversations the original web never anticipated. TechCrunch reported the details.
So the infrastructure layer responds. Microsoft updated Azure services to manage memory sharing between agents and absorb traffic bursts. Databricks and Snowflake reposition their platforms as retrieval systems built for AI memory rather than traditional analytics. Cloudflare, one month before AWS’s announcement, rolled out tools giving agents persistent environments and instant scalability. The company’s earlier Agents Week announcements included Mesh private networking for agents and an Agent Readiness score to help sites measure how prepared they are for machine visitors. Cloudflare Blog.
Forbes examined the broader pattern in mid-May. Companies once focused on making agents mimic human clicks now build software platforms meant for agents from the ground up. Tim Bajarin wrote that this marks another platform shift on the order of mainframe to PC or internet to mobile. The winners, he argued, will design for agents first. Forbes.
Network World captured Cloudflare’s April moves in detail. The company introduced a private networking fabric called Mesh that assigns AI agents, Workers and devices a shared private IP space. It also opened a Registrar API so agents can register domains directly. These pieces form part of what Cloudflare calls the agentic cloud. Network World.
Yet the web itself needs new rules. Sites built for human readers must now offer structured data, clear permission signals and machine-readable summaries. Without them agents scrape inefficiently or trigger blocks. Cloudflare’s Agent Readiness score attempts to quantify that gap. The score, the company hopes, will push publishers to adapt faster than they did when search bots first arrived.
Some builders look backward for answers. A May 18 report noted that Infoblox and GoDaddy back open standards called DNS-AID and Agent Name Service. Both rest on the Domain Name System, the same 1980s protocol that still routes nearly every web request. The idea is simple. Give agents verifiable identities and discoverability without inventing new centralized registries. TechRevolt News.
Interoperability remains the weak point. A LinkedIn analysis compared today’s agent infrastructure to the internet in 1990. Separate systems work well enough in isolation. They fail when asked to coordinate at scale. Open-source efforts such as Project NANDA push for shared protocols so agents from different creators can find, trust and transact with one another. LinkedIn.
BasisSet took the argument further in January. Agents broke existing developer tools. Secure sandboxes, browser automation layers, agent-native databases and dependency management all need rethinking. The firm sees fresh opportunity in the rubble. New infrastructure categories are forming around the specific demands of autonomous code. BasisSet.
McKinsey examined the enterprise view in April. When agents orchestrate work across departments, infrastructure stops being passive support. It becomes active participant, provisioning resources on intent, enforcing governance and optimizing in real time. Data leaders must now build shared foundations before the agents multiply. McKinsey.
The hardware side feels the pressure too. National Bank of Canada analysts argued in late May that agentic AI requires an upgraded “information racetrack.” Optical fiber, networking gear, specialized processors and faster interconnects all face demand spikes. Corning’s recent multi-billion-dollar deals with Meta and Nvidia signal how seriously suppliers take the shift. National Bank.
Gartner’s 2026 predictions, echoed across vendor briefings, forecast that agentic operations will reshape infrastructure and operations teams. By 2030 half of organizations could run autonomous agents handling routine network tasks. The transition from human-led to agent-assisted NetOps is already underway at forward-leaning enterprises.
But risks accumulate. Who bears liability when an agent books the wrong flight, signs an erroneous contract or exposes sensitive data? How do sites defend against agent-driven scraping that overwhelms servers yet follows no clear robot.txt standard? Those questions still lack firm answers. The infrastructure rebuild races ahead of the policy and legal frameworks that should govern it.
Google’s latest Search updates at I/O demonstrated one path forward. The company now generates custom UI components on the fly, fans out research across agents and deploys code within the response itself. Robby Stein, vice president of product for Search, described the process during the keynote. The system plans, researches and builds in one fluid motion. Humans watch results appear. Agents do the work. The Letter Two.
So the pattern repeats. First came search engines that forced sites to publish clean HTML. Then social platforms that demanded shareable cards and real-time updates. Now agents demand structured knowledge, instant scalability, verifiable identity and economic rails built for machines. Each wave forces builders to adapt or fade.
AWS’s OpenSearch move, Cloudflare’s Mesh fabric, Microsoft’s memory-sharing primitives and the quiet standardization efforts around DNS all point the same direction. The internet is no longer a library for humans to browse. It becomes a distributed operating system for autonomous software that acts, transacts and learns at scale.
The rewrite has started. Companies that treat agents as first-class citizens in their architecture will set the standards others follow. Those that wait risk building on foundations already cracking under machine load.
And the machines? They aren’t waiting.
WebProNews is an iEntry Publication