Describing the Shot in Words

Opening Scene

A skilled coach can often correct an archer’s shot dramatically just by describing, precisely and clearly, exactly what needs to change — no retraining required, just better instruction. The archer already has the underlying skill; they simply needed it directed more precisely. This is exactly what a well-crafted prompt does for a language model, and it’s worth taking seriously as a first lever before reaching for anything more involved.

In Plain English

Prompt engineering — the deliberate crafting of instructions, examples, and context, covered fully in this content library’s dedicated series — is usually the right first lever to reach for when a model’s behavior isn’t yet what’s needed. It’s fast to iterate on, fully reversible, and requires no training infrastructure at all. A genuinely large share of behavior problems that look, at first glance, like they need fine-tuning turn out to be solvable through better prompting alone.

The Old Way

Before prompt engineering was recognized as a rigorous, deliberate discipline in its own right, many practitioners underused it as a lever:

Some practitioners reached for fine-tuning prematurely, before genuinely exhausting what better prompting alone could achieve, taking on unnecessary training cost and complexity.
Prompt crafting was sometimes treated as informal trial and error rather than a deliberate, systematic discipline with established, transferable techniques.
There wasn’t yet a well-established norm of exhausting prompting as a first step before considering fine-tuning.

Recognizing prompt engineering as a rigorous first lever, worth genuinely exhausting before considering fine-tuning, reflects real accumulated practical wisdom.

What’s Changing (and Why AI Is the Reason)

Well-established prompt engineering techniques — covered fully in this content library’s dedicated prompt engineering series — increasingly resolve behavior problems that once seemed to require fine-tuning.
Organizations increasingly adopt “exhaust prompting first” as an explicit norm, given how much faster and cheaper it is to iterate on than fine-tuning.
This connects directly to the cost comparison covered later in this series in Article 10 — prompting’s near-zero iteration cost makes it the obvious first thing to try.

The Metaphor, Fully Extended

The Archer	Prompt Engineering as First Lever
A coach correcting a shot through precise description alone	A well-crafted prompt correcting model behavior through precise instruction alone
No retraining required, just clearer instruction	No fine-tuning required, just clearer prompting
Fast to try, fully reversible, no special equipment needed	Fast to iterate, fully reversible, no training infrastructure needed
Many “training problems” turning out to be instruction problems	Many “fine-tuning problems” turning out to be prompting problems

For Beginners: What to Actually Do

Practice genuinely exhausting prompt improvements — clearer instructions, better examples, more explicit formatting — before assuming a problem needs fine-tuning.
Learn the core prompt engineering techniques covered in this content library’s dedicated series, since they’re the concrete toolkit for this first lever.
Get comfortable with the discipline of iterating on a prompt systematically, rather than treating prompt crafting as unstructured trial and error.

For Practitioners and Leaders: The Deeper Layer

Establish “exhaust prompting first” as an explicit organizational norm before approving fine-tuning projects, given the cost asymmetry involved.
Invest in prompt engineering discipline and tooling, since it resolves a genuinely large share of behavior problems without ever requiring fine-tuning.
Track how often fine-tuning requests actually get resolved through better prompting instead, as a useful signal for calibrating this norm over time.

Quick Recap

Prompt engineering is usually the right first lever to reach for when model behavior isn’t yet what’s needed.
It’s fast to iterate on, fully reversible, and requires no training infrastructure.
Many behavior problems that look like they need fine-tuning are actually resolvable through better prompting alone.
Organizations increasingly adopt “exhaust prompting first” as an explicit, deliberate norm.

Where This Fits in the Series

Article 4 covered prompting as the right first lever to try. Article 5 covers what happens when the sights genuinely run out of adjustment, and prompting alone isn’t enough.