GitHub Previews 'Agentic Workflows' for CI/CD

- [Agentic workflows are](https://vertexaisearch.cloud.google.com/grounding-api-redirect/AUZIYQF0LoTW24KN51FI1LLu0F3-0YJY2LF7ba-8DH5Dp80MV_v2CAjUvl6ESt0F5arsmIGnDCQ3QYelWBGIxkYM2OTdgt1pj9O9GaL65i2T1C2fDnOKgJK7u_2_7KqsGiY58Fpsi4zoz2EIEN1r8utIDd8ANB2LKaJHDpSmv-5p9Yl2NrG0mrvSX2avdLlK7PFWm8vdzFu-NbWJwAQr) defined in Markdown files using natural language to describe the desired outcome, which are then compiled into standard GitHub Actions YAML files via a command-line interface. This approach is designed to be more intuitive than writing complex YAML files from scratch. - The feature is designed with a "security-first" model, where agents run in a sandboxed environment with read-only permissions by default. Any actions that modify the repository, such as creating a pull request or labeling an issue, must be explicitly approved through a "safe outputs" mechanism. - This initiative is part of a broader concept GitHub calls "Continuous AI," which integrates AI into the software development lifecycle to augment, not replace, traditional CI/CD pipelines. Agentic workflows are intended for non-deterministic tasks like analyzing CI failures or improving test coverage, while core build and release processes should remain deterministic. - The introduction of agentic patterns is projected to shift the role of software engineers from writing code line-by-line to orchestrating and supervising systems of AI agents that handle implementation details. This allows developers to focus more on system architecture and strategy. - Enterprise adoption of such AI systems necessitates robust AI governance frameworks to manage risks related to compliance, security, and ethics. These frameworks are becoming a key procurement requirement for enterprise buyers, who demand transparency and accountability in how AI is used. - The market for agentic AI is projected to grow significantly, with one Gartner prediction suggesting that over a third of enterprise applications will use AI agents by 2028. This growth reflects a broader trend of moving beyond simple automation to autonomous systems that can reason, plan, and execute complex tasks. - In practice, enterprises are using agentic AI to automate tasks like document review, with JPMorgan Chase's COIN system reportedly performing the equivalent of 360,000 hours of manual work annually. Other applications include predictive maintenance and cybersecurity threat detection. - The underlying technology allows an AI agent to break down a high-level goal into sub-tasks and execute them across multiple steps. This process involves a continuous loop of perception, planning, action, and learning from the outcomes.