Over the past few months, I’ve been learning and building AI Agents / LLM-driven automation workflows. I learned a lot - some exciting, some challenging and I decided to document my learnings in this blog. LLM Workflows and Agentic Automation is Backend Engineering with Extra Steps 🔗Yep! Maybe I’ll ruffle some feathers, but building AI agents is very much a backend / automation pipeline-driven process. The same basic principles apply — and that’s actually a good thing.

One of my first jobs as a college graduate was writing Robotic Process Automation (RPA) workflows using Leapwork. I was able to automate several repetitive, time-consuming tasks, which gave me a deep appreciation for the power of engineering. RPA involves using software “bots” to mimic human interactions with digital systems. These aren’t physical robots, but rather modular software programs designed to perform tasks like data entry or web scraping.

In the programming world, there’s a rising trend some call vibe coding — a term popularized by Andrej Karpathy, meaning you let AI agents write all the code for you, blindly, without personally reviewing or understanding it. You just “vibe,” prompt after prompt, never stopping to check, learn, or correct the output yourself. Let me be clear: I’m not talking about using LLMs, RAG, or prompting to assist coding, generate snippets, or look up APIs.