Software Engineering Is Becoming Plan and Review — Louis Knight-Webb, Vibe Kanban

Key Concepts

• Plan-and-Review Paradigm: The shift in software engineering where human effort moves from writing code to planning tasks and reviewing AI-generated outputs.
• Agentic Workflows: The evolution of AI coding agents from simple code completion (GitHub Copilot) to autonomous execution of complex tasks involving testing, type checking, and tool calling.
• Focus Maxing: A proposed methodology for managing multiple, long-running AI agent streams simultaneously to optimize human productivity.
• Time Horizon: The increasing duration of AI agent tasks, moving from seconds to 5–20+ minute execution windows.
• Vibe Kanban: A tool designed to parallelize agentic workflows, allowing developers to manage multiple streams of work.

---

The Shift in Software Engineering

The speaker, Louie, argues that the role of the software engineer is fundamentally changing. As AI tools like GitHub Copilot, Cursor, and Claude Code have matured, the time spent writing code has significantly decreased. This time has been displaced by two primary activities: planning and reviewing.

The Two Approaches to AI Collaboration

1. The Plan-Based Approach:
   • Methodology: Investing significant time upfront to create comprehensive markdown plans or specs.
   • Benefits: Reduces the need for iterative corrections, minimizes edge-case errors, and results in higher-quality outputs.
   • Recommendation: This is the preferred method for backend development, refactoring, and migrations.
2. The "YOLO" (In-the-Loop) Approach:
   • Methodology: Providing minimal instructions and relying on iterative back-and-forth with the agent.
   • Benefits: Faster to start, but often more time-consuming in the long run due to constant review cycles.
   • Recommendation: Useful for front-end development where stateful interactions and visual edge cases make upfront planning difficult.

Core Takeaway: "Spending 5 minutes of planning saves you 30 minutes of reviewing AI-generated code."

---

Managing Long-Running Agents

As agents become more capable, they perform more complex tasks—such as running type checkers and executing Playwright tests—which increases their execution time.

• The 5-Minute Threshold: Once an agent’s task exceeds 5 minutes, the developer cannot simply "wait" for the output. This necessitates a shift in behavior toward parallelism.
• Focus Maxing: Developers must transition into "managers of multiple streams of work." Instead of deep-diving into one task, they should manage several concurrent agentic processes, reviewing one while another is executing.

---

Future Outlook: The Role of the Human

The speaker identifies the next major breakthrough as AI-driven QA. He anticipates a future where agents can autonomously run projects, click through interfaces, and identify bugs, effectively closing the loop between development and deployment.

The human role will evolve into:

• Task Architect: Defining the scope and logic of the work.
• Reviewer: Ensuring the code meets quality and security standards.
• Shepherd: Managing the deployment process and reacting to feedback/comments automatically.

---

Case Study: The Shutdown of Vibe Kanban

In a live demonstration, the speaker used his own product, Vibe Kanban, to initiate the shutdown of his startup.

• Business Context: Despite having 30,000 monthly active users and 25,000 GitHub stars, the company struggled to monetize.
• The Economic Challenge: The speaker noted that in the current AI market, profitability is largely restricted to those selling to the enterprise or those reselling tokens. Vibe Kanban acted as an interface layer, which proved unsustainable when users spent $30 on the tool but $3,000 on underlying model tokens.
• Lessons Learned:
  • The importance of hiring for enterprise sales early.
  • The value of working with high-caliber teams.
  • The realization that "hard work" is a learned skill that requires team alignment and motivation.

---

Conclusion

The transition to an AI-first development environment is inevitable. Software engineers must move away from the "code-writing" mindset and embrace a "planning and management" framework. By leveraging tools that support parallel agent execution and focusing on high-level task architecture, developers can maintain productivity as AI agents take on increasingly autonomous and time-intensive roles.