Sift agent

AI-assisted development is changing how quickly software gets built. Developers can now generate and ship code faster than ever, resulting in a growing volume of pull requests. But every change still needs to be tested, and CI pipelines are increasingly becoming a bottleneck, slowing teams down and consuming valuable computing resources. Our project tackles this bottleneck with an AI-powered agent that makes GitHub Actions workflows faster and more efficient. Instead of simply recommending improvements, our system analyzes existing workflows, identifies inefficiencies, proposes targeted changes, and runs real experiments comparing the original workflow against the modified version. It then evaluates whether the changes actually improved performance without compromising the existing tests. Most importantly, our agents open real GitHub pull requests containing the proposed workflow changes, experimental results, and supporting evidence. Developers can inspect exactly what was changed, see whether it helped, and decide whether to merge it. Unsuccessful experiments are also documented rather than being presented as improvements. Our working prototype has already tested changes across real open-source repositories and generated reviewable draft pull requests. Our vision is to make CI optimization an autonomous, continuous process so that as AI accelerates software development, the infrastructure behind it can keep up.