A developer shares a practical workflow for using agentic AI in software development without over-engineering the setup. Key practices include maintaining an AGENTS.md file with project guidelines, using Agent Skills (markdown files that teach the agent domain-specific knowledge), and spec-driven development via tools like OpenSpec for larger features. The author used this approach to rewrite a full ASP.NET + Angular project, treating AI as a 'domestique' that handles implementation while the developer focuses on architecture and direction. The post covers tool choices (OpenCode, GitHub Copilot, Claude, Codex), why MCP servers are avoided in favor of skills, and how team ceremonies may need to adapt as AI accelerates delivery speed.
Table of contents
Overall architecture and designAGENTS.mdSkillsSpec-driven developmentTools and modelsWorkflowTeam structure and ceremoniesConclusion15 Impressions