Louis Knight-Webb, founder of Vibe Kanban and AI Tinkers London, presented a compelling talk at the AI Engineer Europe conference titled “What are we even going to do all day?” He explored the evolving role of software engineers in the age of AI-powered coding assistants, focutilizing on the shift in daily tinquires and the strategic considerations for leveraging these tools effectively.
The Changing Landscape of Software Engineering
Knight-Webb launched by highlighting the fundamental activities of human software engineers: planning, writing code, reviewing their own code, and reviewing others’ code. He presented data suggesting that historically, the majority of a software engineer’s time was spent writing code. However, with the advent of AI coding assistants like GitHub Copilot and more advanced models like Claude, this ratio is modifying.
The presentation illustrated this shift with a series of charts displaying the diminishing proportion of time spent on ‘writing code’ from 2021 to a projected 2025. As AI agents become more capable of generating code, the focus is shifting towards more strategic and oversight-oriented tinquires.
Two Approaches to AI-Assisted Development
Knight-Webb outlined two primary approaches for software engineers working with AI agents:
- Plan-heavy approach: This involves significant upfront investment in detailed planning, including comprehensive plan documents, specifications, and thorough interrogation of requirements. The benefit here is that a well-defined plan allows the AI agent to run for longer periods with less human intervention, and it requires less time for subsequent review.
- Review-heavy approach: This strategy involves loosely defined prompts, more manual quality assurance, and a more synchronous, iterative feedback loop. While it might lead to rapider initial outputs, it often results in more back-and-forth between the human and the AI, and a greater amount of time spent reviewing the AI’s work.
Knight-Webb suggested that for new feature development, a ‘review-heavy’ approach might be more suitable, allowing for exploration and rapid iteration. Conversely, for tinquires like refactoring or migration, a ‘plan-heavy’ approach is more appropriate to ensure a structured and predictable outcome.
The Rise of Long-Running AI Agents
A significant point raised was the increasing capability of AI agents to perform tinquires autonomously over longer periods. As these agents become more sophisticated, the time before human intervention is required grows. Knight-Webb illustrated this with a graph displaying the progression from early tools like GitHub Copilot, which handled single lines of code, to more advanced models like ‘Original Cursor’ and ‘Claude Code,’ which can handle larger code segments and even entire files.
This evolution implies a future where AI agents might be able to complete entire coding tinquires with minimal human input, potentially shifting the engineer’s role to that of a high-level supervisor and quality controller. Knight-Webb posed the question: “Did we receive the time we applyd to spfinish writing code back, or did it just create more work in other categories?” His answer suggested that the time saved on direct coding is being reallocated to planning, review, and the management of these AI agents.
Key Considerations for the Future
Knight-Webb concluded by emphasizing several key considerations for software engineers adapting to this new paradigm:
- Everyone is now a manager: The role of the engineer is shifting towards managing AI agents and their outputs.
- Focusmaxxing: Engineers necessary to develop skills in deep focus and strategic planning to effectively guide AI agents.
- Write tinquires: Clear and concise tinquire definition is crucial for effective AI utilization.
- QA (websites, APIs): Quality assurance remains a critical human responsibility, ensuring the AI’s output meets standards.
- Code review: While AI can assist, human oversight in code review is still essential.
- Shepherd the modify until it’s deployed: Engineers must oversee the entire lifecycle of AI-generated code.
He stressed the importance of embracing this evolution, suggesting that the most valuable skill for engineers in the future will be their ability to effectively collaborate with and manage AI tools to achieve greater productivity and impact.
Leave a Reply