Agentic Coding: Hype or Reality?

This post reflects on the author's experience with LLMs and critically assesses the hype surrounding 'agentic coding'. While LLMs can generate usable code, building complete software projects, like an HTTP/2 server, requires intense micromanagement and algorithmic supervision. LLMs frequently get stuck, demanding human intervention and context adjustments. The author argues that current 'agentic coding' tools are largely overhyped, their success relying on the effort of experienced engineers rather than autonomous LLM capabilities. Only by addressing the problem of LLM context management can their true potential be unleashed.