What Agentic AI in the Hands of Actual Programmers Looks Like

ABSTRACT

There is a massive difference between 'bot spam' and 'expert acceleration.' This post explores how real engineers use agentic tools like Claude Code to storm repositories with successful PRs. Rackauckas argues that the difference lies in knowing the codebase and domains—using LLMs as accelerators for tedious work rather than just guessing. Using examples from SciML, he shows how prompts with deep context lead to scaffolding that actually works.