How to Expertly Use LLMs in Development Workflows

A practical framework for using LLMs responsibly in professional software development, contrasting 'vibe coding' (prompting and committing output without understanding) with 'agentic coding' (structured task delegation where the developer remains in control). The authors argue that vibe coding is fine for prototyping but unfit for client work, while agentic coding treats the LLM as a junior developer under close supervision. Key principles include: starting with your own expertise, thinking before prompting, executing one task at a time, and always reviewing outputs methodically. The post also highlights Laravel as a particularly good fit for agentic coding due to its opinionated defaults, and warns that outsourcing too much thinking to LLMs erodes developer skills over time.