Improving LLM workflows feels like putting lipstick on a pig

Wed Aug 26 2026

No matter how I approach agentic programming (or any general use of LLMs), it always feels like I am putting lipstick on a pig. Skills? Yes, they help a bit. Commands? Good idea for more control. Context engineering? It seems like it should solve everything. But it doesn’t. None of it truly solves what I want it to.

I often consider building my own system to capture my workflow, preferences, and style. Then I find one that promises exactly that, and it still cannot deliver the quality and consistency I expect. It seems like the underlying technology (the LLM) is the limiting factor, even with all the scaffolding built around it.

Every time I rework my approach or think of a new way to fix the shortcomings, I get back to that feeling: I am putting lipstick on a pig. I desperately want it to be a majestic stallion, but at the end of the day it is just a pig.