Prompt Engineer or Prompt Tourist?

What we verified before publishing

Fact-check status: passed

The following claims were verified through Perplexity Sonar before the questions were finalised. The host can reference any of them during play:

• Prompt engineering is the practice of designing and refining instructions given to a large language model to improve output quality, accuracy, and relevance.
• Prompt engineering commonly includes few-shot examples, chain-of-thought reasoning, role assignment, and output constraints.
• Context engineering is described as managing what information persists, gets summarized, or gets dropped so an AI agent's context window stays focused and signal-dense.
• Retrieval-augmented generation (RAG) is a technique within context engineering that pulls relevant documents or data into the model's context window before inference.
• Context engineering also covers memory management, tool orchestration, token budget allocation, context compression, and state persistence across multi-turn interactions.
• Modern models typically use distinct system prompts and user prompts, which are combined using model-specific chat templates.
• Using the wrong chat template, even with small errors such as extra newlines, can cause unexpected performance drops.
• Prompt size matters because prompts that exceed a model's maximum token length can truncate instructions or input data and produce lost context or incomprehensible output.
• Elastic says the system prompt can be pushed out of the model's effective attention span as conversation history, tool outputs, and retrieved data accumulate.
• Elastic describes dynamic tool discovery as a way to avoid listing hundreds of tool descriptions in the system prompt and wasting context window space.
• Glean frames prompt engineering as user-facing and context engineering as developer-facing and system-oriented.
• Glean also notes that incorporating external data retrieval through RAG and APIs helps keep AI responses up to date and reliable.