AI Software Development Workflow

Jun 11, 2025

Summary

  • This session provided an in-depth walkthrough of the current and emerging professional workflows for building software with AI, focusing on tools like Cursor, Taskmaster, and custom rules by Ryan Carson.
  • Key workflow changes were discussed, emphasizing a move from manual AI prompting to a more structured, task-directed approach using a PRD and task breakdowns.
  • The meeting featured hands-on demonstrations of Taskmaster and a lighter-weight rules-based approach, discussing their benefits and areas for customization.
  • Upcoming workshops and additional resources for AI-driven development were mentioned, with an invitation for further questions from the audience.

Action Items

  • (No due date – Attendees/Audience): Submit top questions about AI-driven professional coding in the comments for future coverage.
  • (No due date – Workshop Organizer): Notify interested parties about the upcoming workshop on professional AI-driven software development.

Current State of AI-Driven Software Development Workflows

  • The industry has evolved through three major workflow changes: from tab-completion coding (e.g., GitHub Copilot), to agent mode (AI writing most code via prompts in an integrated environment like Cursor), to the emerging “task-directed development.”
  • Task-directed development uses a PRD (Project Requirements Document) as the central source of truth, which is then converted into high-level tasks and subtasks for the AI to execute.
  • While agent mode increased efficiency, it highlighted issues with prompt fatigue and inefficiency, addressed by the new structured workflow.

Demonstration and Comparison of AI Tools

  • Taskmaster:
    • Facilitates the creation of a PRD from a user’s idea, autogenerates organized task lists and subtasks, analyzes task complexity, and automates execution using dependencies and priorities.
    • Stores instructions per task and supports auto-running tasks to minimize manual intervention, but may feel overly complex for users seeking a more lightweight process.
  • Ryan Carson’s Cursor Rules Approach:
    • Relies on three Cursor rules files to mimic the same high-level process (create PRD, generate tasks, process task list) but is lighter and easily customizable.
    • Prompts for clarifying questions, generates markdown checklists for tasks, and marks completed items automatically.
    • Found to be easier to read, edit, and adapt for specific workflows or preferences.

Professionalism and Adaptation

  • Frontloading creativity and decision-making in the PRD phase is emphasized as the main area for expert input, with AI acting as an efficient "junior developer."
  • The process is adaptable; users are encouraged to modify rules and steps to fit their unique project or organizational needs.
  • The workflow supports ongoing oversight, iterative improvement, and maintains opportunities for professional design and development judgment.

Decisions

  • Adoption of Task-Directed Development as the New Norm — The team (or presenter) is moving towards structured, PRD-first, task-driven workflows for professional-grade AI software development because it reduces errors, increases efficiency, and maintains high standards of creativity and oversight.

Open Questions / Follow-Ups

  • How will more complex, multi-layered projects be managed and customized using these workflows?
  • What further customizations and integrations (e.g., specific component libraries or design systems) are most beneficial for individual or team adoption?
  • Details on the upcoming workshop (date, format) to be shared with interested parties.