What it means here Prompt engineering is the deliberate craft of writing and refining what you tell an AI model so it produces the output you want. In this channel it is treated not as a dying novelty but as an enduring skill: precisely describing what you want — the behavior, the style, the details — is the core of building with AI. As Kieran Klaassen puts it, saying exactly what an app should do and how its background or button style should look “is actually… that’s not going away and that is prompt engineering.”[^42a1ae89-412b-5fd8-9efa-752d00aff64c@0] The workflow view Jared Zoneraich frames prompt engineering as iterating on the “source code” of an LLM application through three core primitives: the prompt, the eval, and the data set.[^d351f8cb-2125-5ece-8488-6f6ee66c9ee2@120] Its lasting job, he argues, is putting domain knowledge into your LLM system — the pleasantries may be automated away, but those primitives remain.[^d351f8cb-2125-5ece-8488-6f6ee66c9ee2@151] The channel’s broader takeaway: strong results come less from one clever sentence and more from building a workflow around the prompt, encoding style and domain knowledge into reusable artifacts, and correcting the system when it misses. Why the technique matters Getting good results from AI models is largely about how prompts are crafted — Linus Lee’s experiments with ChatGPT and Notion AI show prompting technique as the lever that shapes what the model returns.[^3e5745ab-279e-5154-9c22-9015dba70a32@0] And because AI makes software dramatically cheaper to build, the question of what you are building matters more than ever — which makes precise description, not trial and error, the skill to explore.[^42a1ae89-412b-5fd8-9efa-752d00aff64c@0] - Precise description is the skill: stating exactly what you want — behavior, style, details — is prompt engineering, and it is not going away. - Three primitives: prompt, eval, and data set form the “source code” you iterate on. - Domain knowledge is the point: prompt engineering endures as the practice of putting what you know into the LLM system. - Workflow beats one-liners: reusable artifacts and correction cycles outperform one-off trial and error.