Topic

From Individual Productivity to Team Productivity: Practical Approaches to Agentic Software Development

Agentic development tools have become widely adopted, yet productivity gains in most teams remain at the individual level: each person has an agent, and the benefits stop there. Using a real-world project as the main case study, this talk explores how we evolved a Coding Agent from an individual tool into an operational software development organization built around task delegation, collaborative execution, ticket-based delivery, acceptance and verification loops, and accumulated experience. Humans shift from writing code to setting direction, validating results, and maintaining boundaries, while Agents handle the implementation. The project uses this very system to develop itself, creating a self-improving development cycle. Outline: 1. Starting Point: The Ceiling of Individual Productivity The real bottlenecks of a single person working with an Agent: context degrades over the course of a session, parallelism is limited, and experience is lost when sessions end. 2. The Key Shift: Treating Agents as an Organization, Not Just Tools The fundamental unit of collaboration: one task → one ticket → one Agent → one pull request (PR). 3. The Human Role Reimagined: Maintainer Agents execute; humans make the decisions. Humans determine what gets merged, when to stop, and which direction to take. Acceptance testing takes place in real-world environments, with trust granted in graduated levels. 4. Organizational Memory: What Outlasts a Session When an Agent finishes a task, it records lessons learned. After human review, these lessons are passed on to the next Agent, ensuring that knowledge is retained rather than lost with individuals or sessions. 5. From Harness to Loop: A Self-Improving Development Cycle Using the system to develop itself. Current project status: more than 30 Agent-implemented PRs are fully merged through the entire workflow in a single week, with one person handling the reviews. Next steps include supporting multiple Coding Agents and further closing the loop on experience feedback. 6. A Transferable Practical Checklist No product pitches—just reusable methods: tickets, acceptance criteria, PR standards, and lessons-learned feedback loops that teams can put into practice immediately. Key Takeaways: 1. A transferable approach to moving from individual Agent usage to organization-wide Agent collaboration, with practices that can be implemented immediately. 2. A redefined understanding of human responsibilities in an Agent-driven development organization, including what must remain under human control and why. 3. Concrete methods for retaining and reusing Agent experience beyond individual sessions. 4. A clear understanding of how to implement a self-improving development cycle, moving beyond vision to practical execution.