Our dev team is looking into "Agentic Workflows" where multiple AI agents handle different roles like QA, Frontend, and Backend. Is it feasible in 2025 to have a "Manager Agent" oversee these to handle Jira tickets and sprint planning? I'm worried about the lack of human nuance in conflict resolution and complex architectural decision-making. Has anyone tried this at scale?
3 answers
We experimented with a multi-agent framework last year for a mid-sized microservices project. We used specialized agents for code review and automated documentation, and while it significantly sped up the CI/CD pipeline, replacing a human PM entirely was impossible. The "Manager Agent" was excellent at tracking velocity and updating status reports, but it failed during "sprint churn" when requirements shifted due to client feedback. AI still struggles with the "why" behind business pivots. It can optimize a path, but it can't always choose the right path when human emotions or office politics are involved in the stakeholders' requirements.
Jennifer, your point about "sprint churn" is interesting. Did you find that the agents were at least helpful in identifying technical debt or dependency bottlenecks before the human PM even noticed them?
I think we are moving toward a "Centaur" model where the PM uses an AI Agent swarm to handle the grunt work of ticket grooming, while the human focuses on the strategy.
I agree, Susan. The "Centaur" approach is definitely the most realistic. At our firm, we’ve seen a 30% reduction in meeting times because the agents summarize all technical blockers beforehand.
Actually, Christopher, that was the agents' strongest suit! They could analyze the entire codebase and identify that a change in the auth-service would break three downstream consumers. They flagged these as "High Risk" in Jira automatically. So, while they didn't replace the PM, they became the PM's most valuable analytical tool for risk assessment and dependency mapping.