4/ Product Thinking They design prompts with real use cases in mind. AI isn’t just for demos. It powers: • Customer support
• Internal tools
• Research agents
• Creative workflows Prompts must work inside real products and systems.
PROMPT ENGINEERING
-
Product Thinking: Designing Prompts for Real Use Cases
By
–
-
Model Fluency: How prompt engineers adapt to LLM quirks
By
–
3/ Model Fluency They’re fluent in LLMs: • GPT-4
• Claude
• Mistral
• Gemini
• HuggingFace OSS models Each behaves differently. Prompt engineers tune to model quirks like a developer tunes to APIs. Different models, different logic. -
Language Design with Proven Frameworks for Reasoning and Output
By
–
2/ Language Design They use proven frameworks — not guesswork. • Chain of Thought
• ReAct (Reason + Act)
• Tree of Thought
• RAG (Retrieval-Augmented Generation)
• Few-shot prompts These guide reasoning, structure, and output control. It’s language as architecture. -
Translation Layer: Prompt Engineering as Strategic Communication
By
–
1/ Translation Layer Prompt engineers translate human goals into precise instructions. They break down abstract tasks into structured input. This gives the model the context it needs to deliver useful, relevant output. Prompting = strategic communication.
-
The Professional Framework Behind Prompt Engineering
By
–
What Prompt Engineers Actually Do (And Why It Matters) It’s one of the most misunderstood roles in AI. Not just “good at ChatGPT.” They design logic, structure, and systems. Here’s the professional framework behind this emerging career
-
Perplexity Assistant Fails to Book China Restaurant for Tuesday
By
–
Perplexity assistant My ask “book china live for 5pm Tuesday” Pplx opened open table with the restaurant searched. But not helpful
-
Model Behavior: Quantification vs Uncertainty Expression
By
–
For example, this should be basis points rather than 20%, but suggesting the model put a number on everything causes it to float to the bottom of the list, versus it impressionistically suggesting "I don't know, could be a firmware issue" in the middle of the first paragraph.
-
AI Model Probability Estimates: Rigor and Custom Instructions
By
–
(I don't find any of the models to be particularly rigorous in what is a 90% versus a 98% or an 85%, but they don't often say 90% when they mean 30%, and so I keep the part of my custom instructions which tells them to provide estimates.)
-
AI as Rubber Duck Debugging Tool Quacks Back
By
–
It might be just "rubber ducking" here but it is a useful form of rubber ducking. The duck now quacks back, and some of its suggestions are things that strike me as plausible, but that wouldn't have occurred to me before several additional debugging steps (at ~a day per).
-
Structuring prompts in phases to avoid LLM confusion
By
–
But yes, LLMS can get confused, but I structured the prompt in phases, so it follows it one by one like a group of smaller prompts