Cloudflare announced the Agent Development Lifecycle (ADLC), a framework the company proposes for dealing with a reality in which AI agents can write code faster than teams can review, deploy, and maintain it. Cloudflare believes that relying on the traditional software development lifecycle model is no longer sufficient to handle the volume of code and the speed at which agents produce it.
The software development lifecycle, commonly known as the SDLC, usually consists of the planning, design, implementation, testing, deployment, maintenance, and retirement stages. According to Cloudflare, implementation was previously the slowest and most expensive stage, but it has become the fastest and least expensive with advances in artificial intelligence. This has increased pressure on the other stages, from code review and pull request management to operating systems in production and handling incidents.
From Code Generation to Managing the Entire Lifecycle
Cloudflare says that many companies currently use agents only during the implementation stage, while humans remain responsible for verifying and integrating the results, deploying and monitoring them, and triaging incoming errors. The company believes this model leaves agents disconnected from important parts of the software development lifecycle, despite improvements in models and their ability to work for longer periods and carry out larger tasks.
Cloudflare treats agents as its customers; they can purchase domains, create temporary accounts, and use the full Cloudflare API. From this perspective, the company says agents need APIs and tools that enable them to manage all stages of the development lifecycle on behalf of customers, rather than merely handling the beginning of the process.
Tools Unveiled by Cloudflare
- @cloudflare/ci: A new way to run continuous integration and continuous deployment across millions of repositories, with the ability to self-correct problems and invoke agents to carry out more complex tasks, using Cloudflare Workflows.
- OpenTelemetry tracing in local development: Providing agents with the same level of observability available in production during local development, through Wrangler and Cloudflare’s Vite plugin.
- Cloudflare Agents and Agent Traces: A new space for monitoring, maintaining, and improving agents, centered on the OpenTelemetry traces generated by agents’ work.
- Enforcing engineering standards with artificial intelligence: A presentation of Cloudflare’s experience applying best practices across its repositories and product and system specifications.
- A software factory for the Astro project: A presentation of an experiment to build systems that automatically triage, reproduce, verify, and fix GitHub issues in a large and growing open-source project.
From SDLC to ADLC
Cloudflare introduces the term ADLC as a model aimed at “software factories,” rather than only traditional programming teams. A software factory, according to the concept the company discusses, is an agent-driven system that can receive inputs such as a production bug, a customer report, or an idea for a new feature, and then handle software development, improvement, deployment, and management tasks with greater autonomy.
This does not mean that the role of humans will disappear; Cloudflare notes that many projects are still constrained by steps involving human intervention, such as directing agents, giving them instructions, monitoring their execution, and applying code review feedback. The goal proposed by the company is to shift human time toward work that requires inspiration, taste, and judgment, such as design, talking with customers, and developing bigger ideas.