Senior Developer Skills in the AI Age: Leveraging Experience for Better Results

Abstraction: Senior-developer practices for guiding AI coding agents

Key points:

• Manuel Kießling introduced Cursor (wrapping Claude Sonnet 3.7, Gemini 2.5, GPT-4o) into his team at Joboo; argues senior engineers' traditional practices are what make AI agents effective.
• Mental model: the agent is "an absolute senior in programming knowledge, but an absolute junior in architectural oversight in your specific context."
• Three measures, all forms of "containment": Well-structured Requirements, Tool-based Guard Rails, and File-based Keyframing.
• Guard rails: give the AI make quality/test commands, static analysis (PHPStan), and curl commands so it validates and self-corrects its own work.
• File-based keyframing: create near-empty stub files (namespaces, class/route names) so the AI infers file organization, naming conventions, and patterns rather than inventing them; he rarely lets Cursor create files itself.
• Green-field case (Python ELK-stack monitor, ~zero prior Python) worked functionally but a HackerNews Python dev (necovek) critiqued the code as junior-quality; author then fed the critique back to improve it — green-field in unknown stacks remains a special case.

Connections: Cursor · Manuel Kiessling · Anthropic · Agentic Coding · AI Coding Assistants · Software Requirements

Source: https://manuel.kiessling.net/2025/03/31/how-seasoned-developers-can-achieve-great-results-with-ai-coding-agents/