Cloudflare has officially completed a major overhaul of its widely read corporate blog, transitioning the platform to EmDash, a specialized content management system built specifically to integrate with Astro and Cloudflare’s edge architecture. Alongside the backend migration, the company rolled out a comprehensive frontend redesign featuring native dark mode support, streamlined navigation, and a modern aesthetic aligned with Cloudflare’s broader ecosystem design language.
The migration, which went live on August 12, represents more than a simple visual refresh. It serves as a real-world stress test for EmDash, a pre-1.0 content management platform, as Cloudflare stepped into its signature role as "Customer Zero"—using its own infrastructure and vendor relationships to validate security, scale, and usability before enterprise clients ever touch the product.
We Are Customer Zero
At Cloudflare, the cultural value of "Customer Zero" is deeply embedded in internal engineering standards, famously documented in the company’s internal Codex. The principle dictates that Cloudflare does not merely build products for external users; it deploys them to run its own massive global infrastructure first.
This philosophy places a heavy burden of proof on teams wishing to utilize external software vendors. Engineers must justify why an external solution is necessary, identify any operational gaps, and determine whether those gaps represent true requirements. By running new platforms through the harshest production environments on earth, Cloudflare forces engineering teams to identify and resolve critical bugs immediately. When the company evaluated limitations with its previous CMS vendor and weighed the launch of EmDash, management recognized that adopting the platform internally was the ultimate validation test.
Customer Zero in Action
When initial migration discussions began, engineering teams focused on two fundamental questions: whether the platform actually worked and whether it could handle enterprise-scale traffic.
Evaluating functionality required running through common administrative and publishing user flows. While EmDash held up well during general usability assessments, the team encountered minor gaps typical of pre-release software. The most significant oversight involved scheduled post functionality, which failed to operate correctly until version 0.19.0 of the CMS. While understandable given the early stage of the platform’s development, discovering such limitations ahead of scheduled publication times underscored the value of rigorous internal testing.
Scalability presented an even greater concern given the unpredictable traffic patterns of the Cloudflare Blog. While normal operating loads hover around 75 requests per second, viral posts or coordinated attention can trigger traffic spikes exceeding 5,000 requests per second. Performance is paramount for a web performance and security enterprise, prompting the team to model extreme scenarios using k6, an open-source performance testing tool.
By configuring rigorous simulations—including sudden bursts of up to 7,000 requests per second—the team evaluated failure rates, latency thresholds, and system checks under heavy load. These tests directly informed the final production architecture, which relies heavily on multiple layers of caching ordered by proximity to the user. This multi-layered approach allows Cloudflare to serve approximately 99.5 percent of static files and 70 percent of total requests directly from cache, dramatically improving frontend performance while shielding the underlying database from excessive load.
Frontend Redesign
The migration also provided an ideal opportunity to harmonize the blog’s visual language with Cloudflare’s updated Kumo design system, establishing consistency across the company’s homepage, developer dashboard, and marketing properties.
Among the most requested features from readers was native support for light and dark modes. The engineering team implemented seamless theme switching tied directly to system preferences alongside an explicit manual toggle, ensuring strict adherence to accessibility guidelines across both visual modes. The updated color palettes and code syntax highlighting now adapt dynamically without sacrificing legibility.
The redesign also addressed long-standing user experience friction points. The previous blog layout placed an email subscription input field in the top-right corner of the page, leading many readers to mistakenly type search queries directly into the subscription box. To resolve this, the design team relocated email sign-ups into a dedicated call-to-action block at the bottom of posts, allowing readers to subscribe naturally after finishing an article. Additionally, new interior sidebar features were introduced, including a dynamic progress-tracking table of contents on the right and simplified social sharing tools on the left to encourage developer community engagement.
Rollout Strategy
Ensuring zero downtime for readers and maintaining a reliable fallback mechanism were non-negotiable requirements for the migration. To achieve this, engineers deployed a proxy Cloudflare Worker capable of intelligently routing traffic between the legacy blog and the new EmDash-powered platform.
The proxy Worker utilized version cookies to direct incoming traffic appropriately while providing an automatic fallback to the legacy blog in the event of any unexpected server errors. Leveraging Cloudflare Workers service bindings, the proxy established a direct worker-to-worker connection, bypassing public hostnames, DNS resolution, TLS handshakes, and outbound HTTP connections to minimize end-user latency.
On launch day, the team executed a phased rollout, starting with just one percent of total traffic before incrementally stepping up to 5 percent, 15 percent, and beyond. This gradual approach allowed engineers to monitor system health under real-world production conditions and catch remaining edge cases without disrupting the broader audience. By the end of the day, 100 percent of traffic had been successfully transitioned to the new platform.
Results
Early performance data confirms that the migration achieved its primary objectives of delivering a faster, more reliable reading experience. Comparative metrics tracking 95th-percentile response latencies revealed a stark improvement over the legacy architecture. While the older system experienced periodic latency spikes under heavy load, the new EmDash setup running on Cloudflare Workers maintains a remarkably flat and consistent response profile, comfortably handling traffic rates up to 850 requests per second with minimal error rates.
The architectural overhaul also expanded accessibility for automated agents through two distinct developments. First, Cloudflare released a new Model Context Protocol server for the Cloudflare Blog, enabling AI agents to search content, retrieve recent posts, and read specific articles via standardized tool integrations. Built using EmDash’s intuitive APIs and AI search endpoints, the MCP server was developed in just a few hours. Second, EmDash itself provides an integrated MCP server for blog authors, allowing them to browse, create, edit, schedule, and delete content using agentic tooling at no additional cost.
The new platform faced its first major operational trial during Agents Week, an internal innovation initiative that saw the publication of 28 new posts over a nine-day period, attracting nearly three million pageviews. The blog Worker performed exceptionally well, serving up to 450 requests per second without incident. Furthermore, Cloudflare’s built-in DDoS protection successfully absorbed a 28,000 request-per-second volumetric attack during the event, maintaining uninterrupted availability for readers. While minor editing quirks and scheduling bugs were identified during the intense publishing schedule, those issues have been communicated to the EmDash team for resolution ahead of upcoming platform milestones.
Cloudflare expressed appreciation to the EmDash development team for their responsiveness and collaboration throughout the migration process, highlighting the success of the Customer Zero methodology as the platform prepares for its upcoming version 1.0 release.
Leave a Reply