More than 10 million developers are now building applications on Cloudflare's infrastructure, a jump from 7.4 million just one quarter earlier. The milestone reflects a deeper transformation already underway: software is no longer written solely by people for people. Increasingly, it is written by developers and AI agents, for both humans and machines, and that shift is reshaping what a cloud platform needs to deliver.
From Serverless Compute to a Full Development Stack
Cloudflare's developer platform began nearly a decade ago with Workers, a serverless compute tool built on a different architectural premise than conventional cloud computing. Rather than running applications on virtual machines or containers, which require heavier resources and slower startup times, Workers executes code in lightweight isolates distributed across Cloudflare's global network. That design choice, made years before the current AI boom, is now proving well suited to a world where software needs to start instantly, scale unpredictably, and run close to the people and systems using it.
Over time, that compute layer expanded into a broader stack spanning storage, databases, AI inference, media handling, real-time communication, and observability tools. Products such as Workers AI, Workers KV, AI Gateway, Durable Objects, D1, and R2 now let developers assemble entire applications - from the database to the user interface to the AI model behind it - without routing between separate vendors and systems.
Why AI Agents Are Changing Internet Traffic
The clearest evidence of this shift is in the numbers Cloudflare tracks directly on its own network. Daily requests from AI agents grew by more than 1,700 percent over the past year, and for the first time, more than half of all Internet traffic observed on the network was not generated by humans at all. That means applications today must be built to serve not just browsers and mobile apps, but also autonomous software making requests continuously, at machine speed, around the clock.
This changes the engineering calculus for developers. An application once designed to handle predictable human traffic patterns - spikes during business hours, quiet periods overnight - now needs to absorb constant, automated demand from agents acting on behalf of users and businesses. Infrastructure that cannot scale instantly or respond with low latency becomes a liability rather than a convenience.
Global Network Scale as a Competitive Foundation
Much of Cloudflare's pitch to developers rests on the physical footprint of its network: more than 340 cities across over 130 countries, interconnection with nearly 14,000 other networks, and close to 600 terabits per second of external network capacity. For developers, that scale translates into a practical advantage - code can run physically near the end user or the requesting agent, reducing the delay that would otherwise accumulate when traffic has to travel long distances to a centralized data center.
That proximity matters more as applications become distributed by default rather than by exception. A single service today might need to respond to a human in one country, an API call from a partner business in another, and an autonomous agent querying it from a cloud region somewhere else entirely - all within the same second. Platforms built on centralized, regionally concentrated infrastructure struggle to meet that demand without added complexity or cost.
What This Means for the Next Phase of Software
The growth in Cloudflare's developer base is a signal of where software construction is heading: toward tools that assume AI involvement at every stage, from writing code to running it to securing it once deployed. As agentic systems take on more of the routine work of software development and operation, the underlying platforms supporting them need to offer speed, global reach, and resilience as default features rather than premium add-ons. That expectation is likely to shape how competing cloud and infrastructure providers position themselves in the years ahead, as the distinction between human-facing and agent-facing applications continues to blur.