Cloudflare AI Engineering Stack: 93% R&D Adoption in 2026

Cloudflare has built and deployed a full agentic AI engineering stack on its own products — achieving 93% adoption across its R&D organization in under a year, processing 241 billion tokens monthly through AI Gateway.

Enterprise IT and security leaders across the CEE region are asking the same question: how do you integrate AI into engineering workflows at scale without sacrificing security, visibility, or code quality? Cloudflare's answer is a fully documented, production-tested architecture built entirely on its own platform — and now available to every customer.

What was announced

Cloudflare published a detailed technical breakdown of the internal AI engineering stack it built and deployed over eleven months. The initiative, led by a cross-functional tiger team called iMARS (Internal MCP Agent/Server Rollout Squad), resulted in 3,683 active internal users — 60% of the company and 93% of the R&D organization — using AI coding tools powered by Cloudflare's own infrastructure.

Key metrics from the last 30 days of operation include 47.95 million AI requests, 20.18 million AI Gateway requests, 241.37 billion tokens routed through AI Gateway, and 51.83 billion tokens processed on Workers AI. Developer velocity reached a record high: the 4-week rolling average of merge requests climbed from approximately 5,600 per week to over 8,700, peaking at 10,952 in the week of March 23 — nearly double the Q4 2025 baseline.

Why this matters for CEE

For CIOs, CISOs, and IT directors in Central and Eastern Europe, this announcement provides a concrete, enterprise-grade blueprint for AI-assisted software development that addresses the most common blockers: security, governance, cost control, and consistency across large engineering teams.

The architecture Cloudflare built is not proprietary internal tooling — it runs entirely on shipping Cloudflare products: AI Gateway, Workers AI, Cloudflare Access, the Agents SDK, Sandbox SDK, and Workflows. Every component is available to Cloudflare customers today. Organizations in the CEE region can replicate this stack to accelerate developer productivity while maintaining Zero Trust security controls, full audit trails, and Zero Data Retention enforcement — critical requirements in regulated industries and markets with strict data residency obligations.

The cost efficiency is also significant. Cloudflare's security agent processes over 7 billion tokens per day using the Kimi K2.5 model on Workers AI — at 77% lower cost than a comparable proprietary frontier model. For CEE enterprises managing infrastructure costs carefully, this represents a material operational advantage.

Technical details

Platform Layer

• Zero Trust authentication: Cloudflare Access handles all identity and policy enforcement; no API keys exist on developer machines.
• Centralized LLM routing: AI Gateway provides a single control point for provider key management, cost tracking, per-user attribution via anonymous UUIDs, and Zero Data Retention controls across all model providers.
• On-platform inference: Workers AI runs open-weight models including Kimi K2.5 (256k context, tool calling, structured outputs) on Cloudflare's global GPU network, eliminating cross-cloud latency.
• MCP Server Portal: A single OAuth flow via Cloudflare Access governs access to 13 production MCP servers exposing 182+ tools across Backstage, GitLab, Jira, Sentry, Elasticsearch, Prometheus, and Google Workspace.
• Code Mode proxying: Portal-level Code Mode collapses all upstream tool definitions into two portal-level tools, eliminating per-server context window overhead — reducing GitLab's 34-tool schema from 15,000 tokens to near zero per request.

Knowledge Layer

• Backstage service catalog: A 16,000+ entity knowledge graph tracking 2,055 services, 228 APIs, 544 systems, 1,302 databases, and 375 teams with full dependency mapping.
• AGENTS.md generation at scale: An automated pipeline processed approximately 3,900 repositories, generating structured context files that tell AI agents the runtime, test commands, conventions, boundaries, and dependencies for each codebase.

Enforcement Layer

• AI Code Reviewer: Every merge request receives automated AI review covering code quality, security, codex compliance, documentation, performance, and release impact — 100% pipeline coverage with 5.47M AI Gateway requests and 24.77B tokens processed in 30 days.
• Engineering Codex: Internal engineering standards are codified as agent-readable rules and cited directly in code review comments, turning organizational standards into an enforceable, automated control.
• Multi-agent orchestration: The review coordinator classifies each MR by risk tier (trivial, lite, or full) and delegates to specialized agents; Workers AI handles approximately 15% of reviewer traffic for documentation tasks.

Softprom and Cloudflare

Softprom is the official distributor of Cloudflare in the CEE region, providing access to the full Cloudflare product portfolio including AI Gateway, Workers AI, Cloudflare Access, Zero Trust Network Access, and the Agents SDK. Organizations looking to implement secure, scalable AI-assisted engineering workflows can engage Softprom for technical consultation, licensing, and deployment support.

From launching this effort to 93% R&D adoption took less than a year — running entirely on the same products Cloudflare ships to customers.

This content was prepared as part of the Softprom DistriFlow project — an automated system for monitoring and adapting vendor news. Original source: original article.