Prompting: When It's Enough

Of the two parameters that guide this whole decision — how much external knowledge your task needs, and how much behavioral adaptation you need — prompt engineering is the right answer specifically when BOTH are low. This course's own simple takeaway is direct about this: prompt engineering is sufficient if you don't have a custom knowledge base and don't want to change the model's behavior.

'Behavior,' in this framing, specifically means the model's vocabulary, writing style, and structural conventions — not just what it knows. A task that needs neither custom knowledge injected nor a change to how the model naturally writes and reasons is exactly the case where a well-crafted prompt, working entirely within the model's existing training, gets you there without any additional infrastructure.

Concrete examples of this zone: general summarization of arbitrary text (the model already knows how to summarize, and you're not asking it to use unfamiliar vocabulary), general Q&A about widely-known topics (the model already has this knowledge from pre-training), classification into general-purpose categories, translation between common languages, or code generation in a widely-used, standard style. None of these need you to inject custom, non-public knowledge, and none of them need the model to write or reason in a way meaningfully different from its default behavior.

Prompting's appeal is exactly its simplicity: no infrastructure (no vector database, no fine-tuning pipeline, no GPU training run), no ongoing maintenance burden, and the fastest possible iteration loop — a prompt change takes effect immediately, with none of the latency RAG's retrieval step adds or the engineering investment fine-tuning requires. This is exactly why prompting should always be the starting point when scoping a new LLM application — reach for RAG or fine-tuning only once you've identified a SPECIFIC gap prompting genuinely can't close, not as a default first move.

from openai import OpenAI

client = OpenAI()

def prompting_is_sufficient(task_description: str) -> str:
    """Illustrating a task squarely in prompting's zone: general
    summarization needs no custom knowledge base and no behavior change —
    the model already knows how to do this well out of the box."""
    resp = client.chat.completions.create(
        model="gpt-4.1",
        messages=[{"role": "user", "content":
            f"Summarize the following in 2 sentences, in plain professional English:\n\n{task_description}"}],
    )
    return resp.choices[0].message.content

# No vector database, no fine-tuning pipeline, no custom vocabulary needed —
# a well-crafted prompt against a general-purpose task is genuinely enough
print(prompting_is_sufficient(
    "Our Q2 revenue grew 15% year over year, driven primarily by enterprise "
    "contract renewals, though churn in the SMB segment partially offset this gain."
))