Cloudflare Unveils 'ADLC' Methodology to Replace SDLC

Cloudflare proposed the concept of the Agent Development Lifecycle (ADLC) to replace the traditional Software Development Lifecycle (SDLC), unveiling a 'Software Factory' vision where agents autonomously perform the entire process from code generation to deployment and operations. While AI coding tools have accelerated the implementation phase, the burden on humans has concentrated in the review, testing, and deployment stages. To address this, Cloudflare provides infrastructure and workflows that allow agents to operate autonomously with greater authority.

7 Essential Conditions for ADLC

Cloudflare defined that the following seven conditions must be met for agents to generate, improve, deploy, and manage software.

• Programmatic: Every task requires an API; ClickOps is not viable.
• Horizontally scalable: Preview environments matching each agent's production are essential.
• Reproducible: Tools are needed to reproduce bugs in specific environments.
• Real-time, push based: Event-based triggers are required, while dashboard polling is unsuitable.
• Atomic: Changes must be independently testable, releasable, and rollbackable.
• Permissioned: An escalation mechanism is mandatory when necessary.
• Self-improving: A mechanism for learning from experience is required.

Cloudflare's Solutions and Tools

To meet these conditions, Cloudflare presents Workflow and Artifacts as core infrastructure.

• @cloudflare/ci: An SDK that moves CI steps to Workflow's step.do(), written in TypeScript instead of YAML, supporting parallel execution and cache reuse.
• Self-healing CI: When a build fails, the AI Healing Agent analyzes the cause and pushes a fix commit. Verified fixes are uploaded to a separate branch to prevent bypassing tests.
• Flue: A TypeScript open agent framework independent of models and deployment environments, where Workflow controls when prompts and context are passed to agents.
• Flagship: A native feature flag service that allows agents to enable features for small groups only, performing production validation and ensuring atomicity.
• Agent Traces: A dashboard visualizing model calls, tool executions, and approval processes in agent sessions, supporting OpenTelemetry-compatible harnesses.

Real-World Applications

• Cloudflare Codex: A system where AI code reviewers and spec reviewers automatically check and enforce internal engineering standards (RFCs). It flagged approximately 230,000 violations and reviewed over 600 design documents, converting tacit knowledge into explicit rules.
• Astro GitHub Issue Triage: An automated triage pipeline operated by the Astro team, which reduced over 200 open issues to about 30, with an expectation of reaching zero within a month. It executes a four-step process of reproduction, diagnosis, verification, and fixing using isolated sub-agents.