Cloudflare solved the infrastructure problem. A different problem remains.
Cloudflare now covers the infrastructure layers an agent needs to run: compute, storage, browsing, inference, email, and developer tooling.
Dynamic Workers, Sandboxes, AI Gateway, and Browser Rendering help teams build and operate agents on one platform. A separate question remains: which business actions should those agents be allowed to take?
The teams building agents on Cloudflare can now move faster than ever. The constraint is no longer time to build. It is time to production.
We can build the agent in days. Getting it past security, compliance and operations is where we lose weeks.
Why fast agents still get blocked before production
Compliance teams, security reviewers and operations leads are not blocking agents because they distrust the infrastructure. They are blocking them because they cannot answer the questions their organizations require before autonomous systems touch production data.
What will this agent decide, and under which conditions?
Can we reproduce a specific decision after the fact?
What happens when the agent hits an edge case or an ambiguous policy?
How do we audit every action the agent took this quarter?
What controls prevent the agent from exceeding its authority?
Without answers to these questions, deployment stalls regardless of how good the infrastructure is. Rippletide makes them answerable.
What Cloudflare Agents Week delivered
Cloudflare built a complete agent infrastructure stack across five categories during Agents Week 2026.
Dynamic Workers: isolate-based runtime for on-the-fly code execution, starting in milliseconds, 100 times faster than containers
Sandboxes: persistent, isolated Linux environments with a shell, filesystem and background processes for agents that need a full operating system
AI Gateway: unified inference layer across 14 or more providers, with Workers binding and multimodal model support
Browser Run: rebuilt on Cloudflare Containers, delivering four times the concurrent browser sessions and 50 percent faster response times
Artifacts: Git-compatible storage built for agents, supporting tens of millions of repositories
The result is a platform where agents can be built, deployed and scaled faster than at any point before.
Cloudflare provides the speed. Rippletide provides the accountability.
Cloudflare makes it fast to build and run agents at scale.
The 10-Day Proof helps reviewers establish whether one action:
Has complete and verified context before execution
Is explicitly authorized under your organization's policies
Respects authority boundaries and escalation rules
Produces a structured, replayable decision preview
Can be reviewed by compliance, security, and operations before approval
Together, the two platforms close the gap between build speed and production readiness.
Example: customer support agent on Cloudflare Workers
An enterprise prepares a customer support agent on Cloudflare. Dynamic Workers handle execution and AI Gateway routes inference across providers.
Without Rippletide:
The agent resolves cases quickly but inconsistently
Case closure decisions vary by region and by model version
High-value cases bypass supervisor review
Operations cannot explain why similar cases receive different outcomes
Compliance blocks further expansion pending an audit framework
With the 10-Day Proof:
Required evidence and policy rules are explicit
Case closure and escalation thresholds are deterministic
Every preview links back to the rule, context, and evidence
Compliance receives a concrete review package before expansion
1. Context Completeness
Scenarios expose when required context is missing and the action should escalate.
2. Explicit Policy Rules
Reviewable rules stay outside the model or provider serving inference through AI Gateway.
3. Authority boundaries
Scenarios test the authority limits for the domain and identify high-risk actions that require human review.
4. Deterministic consistency
The same conditions produce the same decision preview across Workers, Sandboxes, and Browser Rendering.
5. Evidence-Linked Decision Traces
Each preview records the facts used, rules evaluated, authority boundaries checked, and outcome reached.
How the First Cloudflare 10-Day Proof Works
Choose one write-action exposed by the Cloudflare agent
Map its evidence, policy, approval, and exception rules
Build representative and unsafe scenarios
Preview proceed, escalate, and block outcomes
Review each reason and source with the production owner
Package the boundary for sign-off
Expand toward Runtime integration after the Safety Case is proven
What this means for engineering and operations teams
A clear first action boundary for security and operations review
Evidence and exceptions available before deployment approval
Scenarios that can be replayed after a model or provider change
A documented path from offline preview to Runtime expansion
FAQ
Frequently asked questions
Start with one high-risk action exposed by a Cloudflare agent. The 10-Day Proof maps its evidence, policy, approval, and exception boundary, then tests that boundary through offline scenarios and decision previews. Runtime integration follows after the Safety Case is proven.
No. The first review uses representative traces and scenarios offline, so it does not sit in the live execution path. Production Runtime performance is measured on the customer workflow during integration.
Yes. The action boundary is modeled outside the prompt and model provider. The same scenarios can be replayed when the model or routing configuration changes, so reviewers can compare decision previews consistently.
Infrastructure speed is not the bottleneck. Compliance, security and operations teams block deployment when they cannot answer basic questions: what will the agent decide, under which conditions, and how do we audit it after the fact. Rippletide makes those questions answerable before go-live.
From build speed to production confidence
Prove one Cloudflare agent action before deployment
Bring one write-action whose evidence or approval boundary is still unclear. We will map it in a free Risk Review.
Close the gap between build speed and production readiness
Give compliance and security a concrete review package
