How Cloudflare Uses Cloudflare OS: 4,000 Apps Across the Company

Internal Adoption at a Glance

Cloudflare CIO Sam Rhea describes internal Cloudflare OS usage as follows.

• Thousands of employees use it every week: usage spans roles from engineering to sales.
• More than 4,000 apps and tools were created in the latest 30 days: employees created them themselves.
• More than 10,000 hours of manual work saved in sales alone (Cloudflare estimate): examples include territory planning and proposal creation.

Cloudflare has not published a list of all 4,000 items or a category breakdown. The figure of 4,000 cannot be validated through individual case studies. For a more concrete evaluation, focus on the examples in the next section whose process details are public.

Six Public Examples

The official article identifies six examples built by employees with Cloudflare OS. Not all six have detailed case studies. The IT help desk example includes process details, while the others describe only use and impact. The table distinguishes the level of detail available.

What the IT Help Desk Example Shows

This is the most detailed of the six examples. Before adoption, employees exported CSV files from the ticket system, imported them into a spreadsheet, created charts, and opened overnight tickets one by one.

In Cloudflare OS v2, AI first writes the dashboard code. The dashboard then runs the same queries through Gatekeeper. Opening the dashboard does not run AI inference and consumes no tokens; AI is called only when drafting a reply.

How Cloudflare Rolled It Out Company-Wide

Cloudflare did not begin by asking every employee to freely build AI apps. When non-engineers were first given coding AI, the company says it produced many apps looking for a problem to solve.

Cloudflare reversed the approach and followed these five stages. The sequence, rather than the tools alone, is the most transferable lesson.

1. 1

   Create an Email Address for Work People Do Not Want to Do

   Employees emailed tedious tasks to the address. It appeared to reach AI, but a small team handled the work behind the scenes with AI assistance.

2. 2

   Have People Classify Hundreds or Thousands of Requests

   The team identified recurring tasks along with required terminology, procedures, data connections, and output formats.

3. 3

   Turn Repeated Work into Skills

   The team continued handling email until enough common patterns emerged, then made the accumulated context and Skills runnable with one click.

4. 4

   Have AI Build Reusable Apps

   Employees describe work in natural language, and AI writes the code that performs it. Apps can run on demand, on a schedule, or in response to an event.

5. 5

   Scale Through Champions in Each Department

   Cloudflare did not create a large dedicated AI department. Early adopters across regions and teams—including a sales leader in London, a solutions engineer in Texas, and a business development employee in Japan—became champions.

Cloudflare did not roll out 4,000 finished apps. It provided an intake channel for tedious work, reusable Skills, a platform for safe creation and sharing, and local champions as one operating system.

Three Principles from Internal Adoption

1. Company Context Matters More Than the Model

Even a powerful model must be briefed from scratch if it does not know internal terminology, approved procedures, decision criteria, or data locations. Cloudflare OS shares curated knowledge and procedures as Skills so improvements made by one employee can be used by other employees' AI agents.

The differentiator in AI adoption is not only the model selected, but whether the company has structured its work so AI can execute it.

2. AI Builds Tools Instead of Remaining the Tool

In the first version, AI inference ran every time a business Skill executed. Rebuilding the same morning report should not consume large numbers of tokens each day. In the newer version, AI creates an app or workflow once, fixed steps run as code, and AI is called only for decisions.

3. Move Deterministic Work Back into Software

Moving repeated steps back into code reduces operating cost compared with sending everything to AI. In the IT help desk example, opening the dashboard no longer requires AI inference. Inference is metered, so this difference grows over time.

What Other Companies Can Learn

Cloudflare OS is open source and can be deployed to your own Cloudflare account. As of August 2026, however, it is in Early Access and the official documentation notes that some areas remain rough. It is not a finished solution simply because it has been installed.

Before copying the source code, apply the following sequence.

1. Create an intake channel for recurring work
2. Organize approved procedures and internal terminology as knowledge
3. Split AI permissions into the smallest practical units
4. Move deterministic steps into apps and workflows
5. Have champions in each department spread improvements

Many generative AI programs stop after distributing a chat interface and teaching employees how to prompt it. Cloudflare OS points to the next stage: organize company knowledge and permissions so AI can build software for each employee. The competitive advantage is not access to the latest model alone, but the ability to turn company work into tools that AI can operate safely.

For practical setup, see the Deployment Guide. For internal data connections, see the Gatekeeper Guide. For the organizational rollout, see Six Steps for Adoption.

Sources

• How we use AI with Cloudflare OS (Cloudflare CIO's internal adoption record), source for adoption figures, six examples, and the rollout method
• Cloudflare OS announcement
• cloudflare-os (GitHub)