AI agents are reshaping the internet and the cloud needs to keep up
Artificial intelligence agents are changing everything, and we are not just talking about the models themselves, but the entire structure that holds the internet together. The way applications are built, scaled, and operated is being fundamentally challenged by a new kind of software that does not follow any of the rules we learned over the past fifteen years of cloud computing.
The cloud computing we know today was designed for a world where humans access applications, not for a scenario where millions of autonomous agents execute tasks simultaneously, make decisions, and consume resources in ways completely different from any human user.
That is exactly why Cloudflare launched Agents Week, an entire week dedicated to rethinking internet infrastructure for the agent era. The company has always been about helping build a better internet, and this time, that means building for what comes next.
And the timing could not be more relevant.
Every week, new coding agents, autonomous assistants, and tools that operate without constant human intervention keep popping up, while the infrastructure underneath is still trying to shoehorn this new reality into molds created in the smartphone era. The cloud as we know it was born from the last major paradigm shift in technology: the rise of smartphones, which put the internet in everyone’s pocket and transformed what it meant to be online.
In this article, you will understand why this shift runs so deep, what technical and economic challenges are involved, and what Cloudflare is proposing to make agents viable at global scale. 🚀
The internet and the cloud were not built for the AI era
When smartphones put the internet in everyone’s pocket, they did not just add users. They changed the nature of what it meant to be online. Always connected, always expecting an instant response. Applications had to handle an order of magnitude more users, and the infrastructure supporting them had to evolve to keep up with that new volume.
The approach the industry took was straightforward: more users, more copies of your application. As applications grew in complexity, teams broke them into smaller pieces, the famous microservices, so each team could control its own destiny. But the core principle stayed the same: a finite number of applications, each serving many users. Scaling meant more copies.
Kubernetes and containers became the standard. They made it easier to spin up instances, distribute load, and tear down what was no longer needed. In this one-to-many model, a single instance could serve many users, and even if the number of users reached billions, the number of things you had to manage remained finite.
Agents break that logic entirely.
One user, one agent, one task
Unlike every application that came before them, agents are one-to-one. Each agent is a unique instance, serving one user, executing one task. While a traditional application follows the same execution path regardless of who is using it, an agent demands its own execution environment: one where the language model dictates the code path, calls tools dynamically, adjusts its approach, and persists until the task is done.
Think about the difference between a restaurant and a personal chef. A restaurant has a menu, a fixed set of options, and a kitchen optimized to produce them at volume. That is most applications today. An agent is more like a personal chef who asks: what do you want to eat? They might need completely different ingredients, utensils, or techniques each time. You cannot run a personal chef service with the same kitchen setup you would use for a restaurant.
Over the past year, we have watched agents take off, with coding agents leading the way, which is no surprise since developers tend to be the first to adopt new technologies. The way most coding agents work today is by spinning up a container to give the language model what it needs: a filesystem, git, bash, and the ability to run arbitrary binaries.
But coding agents are just the beginning. Tools like Claude Cowork are already making agents accessible to less technical users. When agents move out of the hands of developers and into the hands of everyone — administrative assistants, research analysts, support reps, personal organizers — the scale math gets scary fast. 🤯
The math behind scaling agents for the masses
If the more than 100 million knowledge workers in the United States each used an agentic assistant with about 15% concurrency, you would need capacity for roughly 24 million simultaneous sessions. At 25 to 50 users per CPU, that works out to somewhere between 500 thousand and 1 million server CPUs — just for the US, with one agent per person.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Now imagine each person running multiple agents in parallel. Now imagine the rest of the world, with over 1 billion knowledge workers. We are not a little short on compute. We are orders of magnitude short.
So how do we close that gap?
Infrastructure built for agents
Eight years ago, Cloudflare launched Workers, the beginning of its developer platform and a bet on serverless computing without containers. The motivation at the time was practical: the company needed lightweight compute with no cold-starts for customers who relied on Cloudflare for speed. Built on V8 isolates instead of containers, Workers turned out to be an order of magnitude more efficient — faster to start, cheaper to run, and natively suited for the spin-up, execute, tear-down pattern.
What Cloudflare did not anticipate was how well this model would fit the agent era.
Where containers give each agent a full commercial kitchen — with fixed appliances, walk-in coolers, and everything else, whether the agent needs it or not — isolates give the personal chef exactly the counter space, the burner, and the knife they need for that specific meal. Provisioned in milliseconds. Cleaned up the moment the dish is served.
In a world where we need to support not thousands of long-running applications but billions of ephemeral, single-purpose execution environments, isolates are the right primitive. Each one starts in milliseconds. Each one is securely isolated. And you can run orders of magnitude more of them on the same hardware compared to containers.
A few weeks before Agents Week, Cloudflare took this further with Dynamic Workers in open beta: execution environments created at runtime, on demand. An isolate takes just a few milliseconds to start and uses only a few megabytes of memory. That is roughly 100x faster and up to 100x more memory efficient than a container.
You can spin one up for each request, run a snippet of code, and throw it away — at a scale of millions per second.
For agents to move beyond early adopters and into the hands of everyone, they also need to be affordable. Running each agent in its own container is expensive enough that today’s agentic tools are mostly limited to coding assistants for engineers who can justify the cost. Isolates, by running orders of magnitude more efficiently, are what makes the unit economics viable at the scale agents demand. 💡
The horseless carriage phase
While building the right foundation for the future is essential, we are not there yet. And every paradigm shift has a period where we try to make the new thing work inside the old model. The first cars were called horseless carriages. The first websites were digital brochures. The first mobile apps were shrunken desktop interfaces. We are in that phase right now with agents.
You can see it everywhere.
We are giving agents headless browsers to navigate sites designed for human eyes, when what they actually need are structured protocols like MCP to discover and invoke services directly. Many early MCP servers are thin wrappers around existing REST APIs — same CRUD operations, new protocol — when language models are actually much better at writing code than at making sequential tool calls.
We are using CAPTCHAs and behavioral fingerprinting to verify who is on the other side of a request, when increasingly that entity is an agent acting on someone’s behalf — and the right question is not are you human? but rather which agent are you, who authorized you, and what are you allowed to do?
We are spinning up full containers for agents that only need to make a few API calls and return a result.
None of this is surprising. This is how transitions work.
Building for both worlds at the same time
The internet is always somewhere between two eras. IPv6 is objectively better than IPv4, but dropping IPv4 support would break half the internet. HTTP/2 and HTTP/3 coexist. TLS 1.2 still has not fully given way to 1.3. The better technology exists, the older technology persists, and the job of infrastructure is to bridge both.
Cloudflare has always been in the business of bridging transitions. The shift to agents is no different.
Coding agents genuinely need containers — a filesystem, git, bash, arbitrary binary execution. That is not going away. During Agents Week, Cloudflare’s container-based sandbox environments reached general availability (GA), because the company is committed to making them the best they can be. They are also deepening browser rendering for agents, because there will be a long tail of services that do not speak MCP yet, and agents will still need to interact with them.
But the company is also building what comes next: the isolates, the protocols, and the identity models that agents actually need. The work is making sure nobody has to choose between what works today and what is right for tomorrow.
Security in the model, not around it
If agents are going to handle our professional and personal tasks — reading our emails, operating on our code, interacting with our financial services — then security needs to be built into the execution model, not bolted on as a layer after the fact.
CISOs were the first to confront this. The productivity gains from putting agents in everyone’s hands are real, but today most agent deployments are full of risks: prompt injection, data exfiltration, unauthorized API access, opaque tool usage.
A developer’s vibe-coding agent needs access to repos and deploy pipelines. A company’s customer support agent needs access to internal APIs and user data. In both cases, securing the environment today means stitching together credentials, network policies, and access controls that were never designed for autonomous software.
Cloudflare has been building two platforms in parallel: its developer platform, for people who build applications, and its zero trust platform, for organizations that need to secure access. For a while, these platforms served distinct audiences. But the questions how do I build this agent? and how do I make sure it is secure? are increasingly becoming the same question. The company is bringing these platforms together so that everything is native to the way agents run, not a separate layer you bolt on afterward. 🔐
Agents that follow the rules
There is another dimension to the agent era that goes beyond compute and security: economics and governance.
When agents interact with the internet on our behalf — reading articles, consuming APIs, accessing services — there needs to be a way for the people and organizations that create that content and operate those services to set terms and get paid. Today, the web’s economic model is built around human attention: ads, paywalls, subscriptions.
Agents do not have attention in the human sense. They do not see ads. They do not click on cookie banners.
If we want an internet where agents can operate freely and where publishers, content creators, and service providers are fairly compensated, we need new infrastructure for that. Cloudflare is building tools that make it easier for publishers and content owners to define and enforce policies on how agents interact with their material.
Last year, alongside Coinbase, Cloudflare co-founded the x402 Foundation, an open and neutral standard that revives the long-dormant HTTP 402 status code to give agents a native way to pay for the services and content they consume. Agent identity, authorization, payment, and security: all of this needs open standards that no single company can define alone.
Building a better internet has always meant making sure it works for everyone — not just those who build the technology, but the people whose work and creativity make the internet worth using in the first place. That does not change in the agent era. It becomes even more important. 🌍
The real cost of running agents in production
There is a conversation that happens a lot behind the scenes of artificial intelligence product development and rarely shows up in official announcements: the cost of running agents in production can be surprisingly high when the infrastructure was not designed for this type of workload. It is not uncommon to see engineering teams blindsided by cloud bills way larger than expected after putting their first agents into real operation, even with relatively modest usage volumes.
The problem has multiple layers. The first is execution time: agents that do chained reasoning, consult multiple sources, and make complex decisions can run for minutes instead of milliseconds, which completely flips the pricing logic of traditional serverless functions that were optimized for ultra-short executions. The second layer is data traffic: agents that access knowledge bases, call external APIs, and communicate with other agents generate far more data transfer than a conventional web application for the same number of end users.
Cloudflare’s answer to this problem is an execution model based on isolates, which are dramatically cheaper to operate because they consume only the resources strictly needed for each task. The hibernation periods of a Durable Object, for example, do not consume active compute resources, which represents significant savings for agents that spend most of their time waiting for responses from third-party APIs, external language models, or human actions within a hybrid workflow. 💰
A platform for developers and agents
Cloudflare’s vision for its developer platform has always been to offer a complete platform that just works: from experiment to MVP, all the way to scaling for millions of users. But providing the primitives is only part of the equation. A great platform also needs to think about how everything works together and how it integrates into the development workflow.
That work is evolving. It used to be purely about developer experience, making it easy for humans to build, test, and deploy. Increasingly, it is also about helping agents help humans, and making the platform work not only for the people building agents but for the agents themselves. Can an agent find the most up-to-date best practices? How easily can it discover and invoke the tools and CLIs it needs? How smooth is the transition from writing code to deploying?
During Agents Week, Cloudflare is delivering improvements on both fronts — making the platform better for the humans who build on it and for the agents that run on it.
Building the future is a team effort
Building the future is not something anyone does alone. Every major internet transition — from HTTP/1.1 to HTTP/2 and HTTP/3, from TLS 1.2 to 1.3 — required the industry to converge on shared standards. The shift to agents will be no different.
Cloudflare has a long track record of contributing to and helping push the standards that make the internet work. The company has been involved with the IETF for over a decade, helping develop and deploy protocols like QUIC, TLS 1.3, and Encrypted Client Hello. It was a founding member of WinterTC, the ECMA technical committee for JavaScript runtime interoperability. And it open-sourced the Workers runtime.
The company is bringing the same approach to the agentic era. It is part of the Linux Foundation and the AAIF, and is helping support and advance standards like MCP that will be critical to the future of agents. Since Anthropic introduced MCP, Cloudflare has worked closely with them to build the infrastructure for remote MCP servers, open-sourced its own implementations, and invested in making the protocol practical at scale.
What changes for those building with AI right now
If you are building something with AI agents today, whether it is an autonomous assistant, an automation pipeline, or a tool that combines multiple models, the main message from Agents Week is that the available infrastructure is finally starting to align with the real needs of these systems. That does not mean all the problems have been solved at once, but it does mean there are now more suitable primitives to work with, instead of trying to adapt solutions that were created for a completely different purpose.
The adoption of the MCP protocol as a standard for communication between agents and tools is particularly relevant because it reduces ecosystem fragmentation. When every tool has its own way of being accessed by an agent, the result is brittle integration code that is hard to maintain and full of edge cases. An open standard changes that, and Cloudflare, by natively adopting MCP on its platform, is signaling that it is betting on consolidating this ecosystem rather than trying to create a proprietary standard that locks developers into its own infrastructure.
Cloud computing is, in practice, going through a deep conceptual overhaul. The request-response model that dominated the last decade and a half is giving way to a model based on agents that persist, reason, and act continuously. And the infrastructure that will support this new model needs to be designed from the ground up for this purpose, with real isolation, efficient execution, and costs proportional to actual usage.
The internet was not built for AI. The cloud was not built for agents. But Cloudflare has always been about helping build a better internet — and what better means changes with each era. This is the era of agents, and Agents Week is a clear bet that this future arrived sooner than a lot of people expected. 🌐
