Agents have made CI the bottleneck. Faster pipelines are the wrong fix.
Three posts landed in September that I think engineering leaders should read together. Anthropic’s engineering team wrote that their continuous The post Agents have made CI the bottleneck. Faster pipelines are the wrong fix. appeared first on The New Stack.
The article discusses how the rise of AI coding agents has made continuous integration (CI) pipelines a bottleneck in software development. It argues that simply making pipelines faster is not the solution and highlights the need for a shift in how code is verified within the development process.
What happened
The article outlines how AI coding agents have caused a surge in CI job volume, leading to bottlenecks. It highlights that traditional CI pipelines, which were designed for human developers, are not effective for agent-generated code. The author argues that verification should occur within the agent's loop, against the real system, rather than just the repository. The article also proposes solutions such as multiplexing environments and governed verification processes to address the issue.
Why it matters
This is important for AI developers, DevOps professionals, and system architects because it highlights a critical challenge in integrating AI agents into software development workflows. The article emphasizes the need for a shift in verification practices to ensure code works with the entire system, not just the repository, and provides actionable solutions to address this issue.
- Traditional CI pipelines are not suitable for AI-generated code due to increased volume and context loss.
- Verification should occur within the agent's loop, against the real system, not just the repository.
- Multiplexing environments and governed verification processes are essential for system-level verification.
- The goal is to ensure code works with the entire system, not just the repository.
- The future of CI lies in integrating verification into the agent's loop, not just making pipelines faster.
The bigger picture
The shift from repository-based verification to system-level verification can significantly reduce cross-service breakage in distributed systems. It also enables more efficient and reliable development workflows by catching issues early in the agent's loop.
The article concludes that the rise of AI coding agents has fundamentally changed the nature of CI pipelines. The solution lies in shifting verification to occur within the agent's loop, against the real system, rather than just the repository. This approach ensures code works with the entire system and reduces cross-service breakage.
Technical details
- Continuous Integration (CI) pipelines are being overwhelmed by the volume of jobs generated by AI agents.
- Agents write code, open PRs, and wait for CI results, but the context is lost by the time feedback is received.
- Verification should occur within the agent's loop, against the real system, not just the repository.
- Multiplexing environments using Kubernetes clusters allows for shared stable versions of services and lightweight test environments.
- Governed verification processes involve platform teams defining steps for agents to execute against live systems.
FAQ
Why are CI pipelines becoming a bottleneck with AI agents?
AI agents generate a large volume of code and PRs, overwhelming traditional CI pipelines designed for human developers.
What is the main issue with current CI verification practices?
Current CI pipelines verify code against the repository, not the real system, leading to cross-service breakage in distributed systems.
How can verification be improved for AI-generated code?
Verification should occur within the agent's loop, against the real system, using multiplexed environments and governed verification processes.
What is multiplexing in the context of CI pipelines?
Multiplexing involves using a shared Kubernetes cluster with stable service versions and lightweight test environments to support multiple agents efficiently.
Why is governed verification important?
Governed verification ensures that agents follow structured steps to test against live systems, maintaining consistency and safety in shared environments.
This summary and analysis were generated with AI from the original article and may contain errors. Read the original for the full story.
Read the full story
Originally published by The New Stack - AI on Oct 4, 2026.
Comments
Sign in to join the discussion.