Fastly's AI Traffic Surge and What It Rewrites

Here is a number worth sitting with: AI traffic on Fastly's global network grew approximately 30% in the first five months of this year—about 6.5 times faster than human traffic over the same period, according to Fastly's own research.

That's not a projection. That's a measurement of what's already happening on the network—a ratio that tells you something real about which direction the internet is walking.

The Fastly Numbers

Fastly sits in a useful position for this kind of data. Its edge cloud platform processes requests from a wide cross-section of the internet before they hit origin servers, which makes it a reasonable barometer for traffic composition at scale. When the company says autonomous systems are nearing half of all internet requests and reshaping web infrastructure, that's not a marketing projection—it's a reading from the instruments.

IT Brief notes the strain this creates on origin systems: servers that were sized and configured assuming a certain ratio of human-to-machine traffic are now receiving a fundamentally different mix. The architecture assumptions underneath a lot of web infrastructure are quietly becoming wrong.

Pulse2 reports that Fastly is specifically positioning its platform for what it calls "agent commerce"—AI agents transacting on behalf of users, completing purchases, filling forms, executing multi-step workflows without a human hand touching the keyboard. The Skyfire integration the outlet covers brings payment and identity infrastructure to the edge, which is the technical prerequisite for autonomous agents doing anything with real-world consequences.

That last part is where the story gets complicated for anyone tracking it beyond the investor deck. The traffic surge is real, the infrastructure investment is real, and the business case is coherent. And the workers whose jobs sit behind those business operations—the customer service agents, the data reviewers, the content moderators, the operations coordinators who have been the actual error-correction layer inside systems that looked autonomous but weren't—are already being counted against this growth, not alongside it.

What "Agentic" Actually Means at Ground Level

The term "agentic AI" has been doing a lot of work in tech coverage lately, mostly as a synonym for "more capable." But the Fastly data points to something more specific: AI systems that initiate requests rather than respond to them, that browse, query, and transact independently. Yahoo Finance's coverage frames this as machine-driven requests "becoming a key layer of internet operations"—language that accurately describes both the technical reality and its implications for the humans those machines are replacing.

The distinction matters because human traffic and agentic traffic aren't just different in volume—they're different in kind. A human browsing a healthcare portal or filing an insurance claim brings judgment, contextual knowledge, and the ability to notice when something is wrong and stop. An agent completing the same task executes its instructions. The error-correction that used to live in the human step doesn't disappear; it gets relocated—either upstream into the model, downstream into a verification system, or nowhere, which is also an option companies have been known to choose.

This is not an argument against AI agents. It's an observation about what gets obscured when the infrastructure story is told purely as a capacity story.

The Investment Thesis and Its Gaps

Seeking Alpha's analysis treats the traffic surge as straightforwardly bullish for FSLY stock—accelerating growth rates, a platform well-positioned for the agentic web, a buy case built on Fastly's ability to handle machine traffic at scale. The argument is not wrong on its own terms. If AI agents become the dominant mode of internet activity, the companies that sit between those agents and the services they're querying are in a structurally interesting position.

But the investment thesis has a quiet gap: it prices the traffic and doesn't price the displacement. The workers who were, until recently, the humans in the loop—who were reviewing AI outputs, flagging errors, handling the edge cases that agents misread—those workers don't show up as a cost on Fastly's network. They show up as a cost on someone else's income statement, or on an unemployment filing, or not at all. The efficiency gain that drives agentic traffic growth is exactly the loss that doesn't appear in the infrastructure company's reporting.

None of this is unique to Fastly. It's the standard structure of the AI build-out story: the companies building and routing the infrastructure capture and report the upside; the displacement concentrates elsewhere and goes largely unmeasured. What makes the Fastly data interesting is that it makes the substitution visible in a way that most AI coverage doesn't. A 6.5x growth differential between machine and human traffic isn't just a business metric—it's a ratio describing how quickly one kind of internet activity is replacing another.

The Origin Strain Problem

IT Brief's reporting on origin system strain is worth dwelling on, because it gestures at a dynamic that extends beyond server architecture. Origin systems—the databases, APIs, and backend infrastructure that actually hold and process information—were built assuming humans on the other end. Response times, rate limits, authentication flows, error handling: all of it calibrated for human-paced interaction.

Agentic AI doesn't interact at human pace. It queries in parallel, at scale, continuously. The "heavier strain" IT Brief describes is what happens when the assumptions underneath a system stop matching the reality of what's hitting it. That's a solvable infrastructure problem, and Fastly's business is partly built on solving it.

The less-discussed version of that same problem applies to the workers whose jobs were also calibrated for human-paced interaction. Customer service workflows, review queues, operations pipelines—these were designed around a certain volume of exceptions, a certain rate of human judgment being required. When agents handle the routine and leave only the hardest edge cases for humans, the job doesn't just shrink; it changes shape in ways that typically aren't accounted for in how those roles are staffed, compensated, or even described.

What the Data Doesn't Count

Fastly's own blog post frames the traffic growth in terms of "new business challenges and opportunities"—a phrase that accurately names both directions without specifying who faces which. For Fastly's customers building on the platform, the opportunity is real: cheaper, faster, scalable automation of tasks that previously required human coordination. The challenge, named politely in the infrastructure context, is origin strain and security complexity.

The business challenge that goes unnamed is the one faced by workers who spent years learning to do precisely the tasks that agentic systems are now handling. That knowledge—what a valid insurance claim looks like, how a customer's account history shapes what they actually need, when a data anomaly is worth escalating—was the value those workers brought. It wasn't entered into any system. It lived in the people.

When an agentic system takes that job, the worker doesn't just lose the position. They lose their legibility as a cost that mattered—their labor dissolves into the efficiency calculation that made the automation case, and they don't appear again in any of the reporting that tracks how the infrastructure build-out is going. The 6.5x growth ratio is measured. The human it replaced is not.

---

Carmen Rodriguez covers labor, workplace organizing, and worker rights for Buzzrag.