Keeping Humans Accountable as AI Agents Take On the SDLC

Ming Wu, leader of engineering for Dev AI at Atlassian, asserts responsibility for AI agents completing development tasks lies with people. She differentiates the capacity to automate tasks from the duty to comprehend and manage their outcomes. This distinction becomes crucial as teams shift from individual coding prompts to entrusting agents with broader portions of the software development lifecycle.

Wu introduces the concept of governed agent loops, a combination of governance and loops. Governance offers visibility, safeguards, and enforcement; loops enable agents to handle batches of jobs under set conditions, replacing manual task triggering. These loops can encompass an issue, executing the work, and evaluating a pull request.

Even as automation handles more steps, teams must still establish a method to track activity and evaluate the output's alignment with expectations. Wu highlights Jira as a suitable interface for this approach, as it already logs work across teams and organizational boards. She also discusses support for third-party coding agents and a shared context layer to provide agents with a consistent understanding of team priorities, organizational requirements, and business intent.

Coordinating these inputs is a key engineering challenge, especially when multiple agents work on the same project. Wu notes that customers are seeking ways to scale these workflows and prove real improvements in delivery speed. Some software teams are already testing these methods, while traditional industries are at varying stages of adoption.

Wu emphasizes the importance of visibility into work alongside automation, ensuring teams can see what agents are doing, maintain accountability, and gauge if the investment yields beneficial results. While completing more automated tasks is part of this assessment, understanding their impact on the software lifecycle is a more complex issue.