Massive AI Adoption Across R&D
Cloudflare Inc. announced today that 93% of its R&D organization has adopted AI coding tools over the past 30 days—tools built entirely on the company’s own platform. The surge includes 3,683 internal users actively using agentic AI, representing 60% of all employees.
The company processed 47.95 million AI requests in that period, with 295 teams now leveraging AI coding assistants. These numbers, shared exclusively with press, highlight a dramatic acceleration in developer velocity.
“We’ve never seen a quarter-to-quarter increase in merge requests to this degree,” a Cloudflare spokesperson told reporters. The four-week rolling average of merge requests jumped from roughly 5,600 per week to over 8,700, hitting a peak of 10,952 in the week of March 23—nearly double the Q4 baseline.
The AI Engineering Stack: Built on Cloudflare’s Own Products
The initiative, code-named iMARS (Internal MCP Agent/Server Rollout Squad), began eleven months ago with a mandate to truly integrate AI into Cloudflare’s engineering workflow. The tiger team—pulled from across the company—designed a layered architecture where every component runs on a shipping Cloudflare product.
- Zero Trust authentication via Cloudflare Access
- Centralized LLM routing, cost tracking, BYOK, and Zero Data Retention through AI Gateway
- On-platform inference with open-weight models using Workers AI
- MCP Server Portal with single OAuth (Workers + Access)
- AI Code Reviewer CI integration (Workers + AI Gateway)
- Sandboxed execution for agent-generated code via Dynamic Workers
- Stateful, long-running agent sessions using Agents SDK (McpAgent, Durable Objects)
- Isolated environments for cloning, building, and testing via Sandbox SDK (now GA)
- Durable multi-step workflows powered by Workflows—scaled 10x during Agents Week
- 16K+ entity knowledge graph built on Backstage (open source)
“We quickly realized MCP servers alone weren’t enough,” the spokesperson explained. “We had to rethink how standards are codified, how code gets reviewed, how engineers onboard, and how changes propagate across thousands of repos.”
The resulting stack—publicly detailed for the first time during Cloudflare’s Agents Week—is not internal-only infrastructure. Every component except Backstage is a shipping product that customers can deploy today.
Background: From Pilot to Platform
Cloudflare’s AI integration began as an experiment to improve internal developer productivity. The iMARS team—short for Internal MCP Agent/Server Rollout Squad—was formed to build internal MCP servers, an access layer, and AI tooling that agentic systems could actually use at scale.
After the initial sprint, sustained ownership passed to the Dev Productivity team, which already managed CI/CD, build systems, and automation. Over eleven months, the team iterated from basic MCP servers to a full-stack AI engineering platform.
Key milestones included enabling AI Gateway for cost tracking and zero data retention, deploying Workers AI for inference, and integrating Sandbox SDK for secure code execution. The result: a self-reinforcing cycle where internal usage drives product improvements and vice versa.
What This Means for Developer Velocity and the Industry
The doubling of merge requests in a single quarter signals a step-change in engineering productivity, according to Cloudflare. By dogfooding its own platform, the company validates that AI coding tools can significantly accelerate software delivery—without compromising security or compliance.
“This is the blueprint for how enterprises should approach AI in engineering,” the spokesperson said. “Start with strong foundations—Zero Trust, cost control, data retention policies—then layer on agentic AI where it creates the most leverage.”
For the broader tech industry, Cloudflare’s internal numbers offer rare, concrete evidence of AI’s impact on developer productivity. The company’s decision to ship the same tools it uses internally means customers can replicate this stack immediately.
Analysts expect other large engineering organizations to follow suit. “Cloudflare is showing that AI agents aren’t a distant future—they’re already doubling throughput for teams that invest in the right infrastructure,” the spokesperson added.