prompt engineering is dead. long live prompt engineering

most people still comment "prompts" to get the playbooks, when they should probably think a lot more about context engineering!

i always considered llms to be really good translators. not just from one language to another, but also translating ideas into code and data into insights.

with that angle in mind, it becomes obvious that what data you put into an llm is much more important than how you prompt it

so keep the context as slim as possible, prefilter aggressively and feed your llm a clean diet!