By David Liew
The Key Ingredients of Success
As startup founders navigating an AI-powered landscape, we now have access to a suite of "co-founder experts" at our fingertips. By strategically crafting our AI prompts with Roles, Examples, Context, Intent, Presentation, and Evaluation (R.E.C.I.P.E.), we can effectively summon our own virtual CTO, CFO, or legal consultant at a fraction of the cost of a permanent hire. The R.E.C.I.P.E. framework, along with its streamlined counterpart RICE (Role, Intent, Context, Examples), offers founders key ingredients for maximizing the potential of Large Language Models (LLMs) at their disposal.
Cooking Up a Wonderful Experience
Think of a prompt as a recipe - just as a good recipe has essential components that work together to create a delicious meal, an effective prompt has 6 key ingredients:
RECIPE: Role, Examples, Context, Intent, Presentation, Evaluation
Here's a chicken rice example:
Role
Who's cooking?
Prompt: "Act as a Singaporean chef specializing in traditional hawker cuisine."
Examples
Show me what success looks like
Prompt: "For instance, describe the texture like: 'The chicken should be silky-smooth with gelatinous skin, while the rice grains remain distinct yet infused with flavor.'"
Context
What's in your kitchen?
Prompt: "I have access to fresh chicken, pandan leaves, ginger, and garlic, but limited experience with Southeast Asian cooking techniques."
Intent
What are you cooking?
Prompt: "Guide me through making authentic Hainanese chicken rice with all three essential components: the chicken, flavored rice, and dipping sauces."
Presentation
How should it be served?
Prompt: "Present instructions in three sections (chicken, rice, sauces) with clear timing to ensure everything finishes together."
Evaluation
How will you know it's perfect?
Prompt: "The instructions should ensure tender chicken with gelatinous skin and rice that's fragrant without being greasy."
The R.E.C.I.P.E. framework transforms prompt writing from guesswork into a systematic process, ensuring each component harmonizes with the others to produce consistently excellent results.
Here's a starter template to improve your prompting journey:
## Role
Act as an expert prompt engineer who balances teaching with constructive criticism. Analyze my prompt for strengths and weaknesses.
## Examples
Strong prompts typically:
Begin with clear problem statements
Provide necessary background without overwhelming
Include specific formatting instructions
Avoid:
Contradictory instructions
Overly complex nested requirements
Ambiguous objectives
## Context
My prompt engineering skill level is [beginner/intermediate/advanced]. I'm interested in using AI for [specific domain] and struggling with [specific challenge, e.g., "consistent formatting"].
## Intent
Analyze this prompt I've created and provide actionable feedback to improve my prompt engineering skills:
[Insert your prompt here]
## Presentation
Structure your response as:
1. Overall Assessment (2-3 sentences)
2. Component Analysis (each RECIPE element)
3. Suggested Revisions (specific rewrites)
4. Enhanced Version (complete rewrite)
5. Key Takeaways (3 principles to apply)
Use clear language with markdown formatting.
## Evaluation
A successful response will identify specific strengths/weaknesses, provide actionable improvements, teach transferable principles, and leave me with a significantly improved prompt.Does order matter?
The most important elements should either come first (primacy) or last (recency) as these positions receive the most attention. For instance:
Intent-first works well when your request is straightforward: "Help me debug this Python code that's throwing an error..."
Role-first works well when expertise matters most: "Act as an experienced cybersecurity expert. I need to understand..."
Context-first works well for nuanced situations: "I'm planning a wedding for 200 guests with a $15,000 budget. The venue requires..."
Following R.E.C.I.P.E. Order (Role First)
Like the "mise en place" approach where you set up your cooking station before beginning:
Starting with Role establishes the expertise framework first
Great for creative tasks, brainstorming, or when expertise is crucial
Example: Writing marketing copy, creative fiction, specialized analysis
Leading with Intent (I First)
Like stating what dish you're making before gathering ingredients:
Immediately focuses the AI on your specific goal
Ideal for direct problem-solving or straightforward requests
Example: Technical troubleshooting, specific research questions, direct tasks
Starting with Context (C First)
Like describing your kitchen constraints before deciding on a recipe:
Sets up the situation before stating what you need
Works well for complex scenarios with important background details
Example: Business case analysis, situation-specific advice
If in a rush, fry RICE
I made my own RICE heuristic to complement my RECIPE framework. The average task might not need such explicit instructions to obtain a decent result.
RICE: Role, Intent, Context, Examples
RICE is like a perfect stir-fry – quick to prepare but still delicious and satisfying:
Covers the essential ingredients (80% of what makes prompts effective)
Omits the "garnishes" (Presentation and Evaluation can often be implicit)
Faster to compose while still getting excellent results
Role
Who's the expert?
Prompt: "Act as a pediatric nutritionist."
Intent
What exactly do you need?
Prompt: "Create a one-week meal plan for a picky 4-year-old who only eats beige foods."
Context
What's the relevant situation?
Prompt: "The child refuses vegetables but needs balanced nutrition. Parents have limited cooking time on weekdays."
Examples
What does good look like?
Prompt: "For instance, suggestions like 'whole grain pancakes shaped like teddy bears with hidden butternut squash' would be ideal."
Here's a condensed version of the prompt template from above.
## Role
Act as an expert prompt engineer who balances teaching with constructive criticism.
##Intent
Analyze this prompt I've created and provide specific, actionable feedback to improve my prompting skills:
[Insert your prompt here]
##Context
My prompt engineering skill level is [beginner/intermediate/advanced]. I'm interested in using AI for [specific domain] and struggling with [specific challenge]. I understand effective prompts need clear objectives, relevant context, formatting instructions, examples, roles, and evaluation criteria.
## Examples
A helpful analysis would:
1. Identify specific strengths and weaknesses
2. an enhanced version of my prompt
3. Share 3 key principles I can apply to future promptsThe Length vs. Specificity Tradeoff
This is perhaps the most important meta-consideration in prompt engineering. It's like the difference between a concise recipe card and a detailed cookbook:
Short Prompts:
✅ Less overhead, faster to write
✅ More room for Claude's creativity
✅ Often better for creative or exploratory tasks
❌ May lead to more variability in outputs
❌ Requires more inference from Claude
Detailed Prompts:
✅ More consistent, predictable results
✅ Better for technical or precise tasks
✅ Reduces need for clarification or refinement
❌ Take longer to craft
❌ May constrain Claude's helpful insights
❌ Can introduce conflicting instructions if not carefully composed
Think of it like a GPS navigation system: a short prompt might just say "Take me to downtown," while a detailed prompt specifies "Take the highway to exit 23, then use the second lane to turn right onto Main Street, avoiding construction on 5th Avenue."
The Specificity Sweet Spot
The ideal prompt strikes a balance - like a well-engineered bridge that's neither overbuilt nor underbuilt.
Conclusion: You can improve your interaction with LLMs.
Consider the Key Ingredients of Success:
RECIPE = Role, Examples, Context, Intent, Presentation, Evaluation
RICE = Role, Intent, Context, Examples
Order matters, so tweak the ingredients accordingly.
Intent-first works well when your request is straightforward: "Help me debug this Python code that's throwing an error..."
Role-first works well when expertise matters most: "Act as an experienced cybersecurity expert. I need to understand..."
Context-first works well for nuanced situations: "I'm planning a wedding for 200 guests with a $15,000 budget. The venue requires..."
Have fun using the "RECIPE" and "RICE" techniques to improve your interaction with LLMs!
About David Liew
I'm in an exciting season of building and exploring - while my path might seem non-linear, I've never been more confident about the intersection of skills and experiences I bring to the table. If you're interested in discussing AI development, strategic technology implementation, or creating systems that help people thrive in our rapidly changing world, I'd love to connect over coffee!
You can find me on LinkedIn, where I actively share insights about technology, personal growth, and the future of work. For a more complete picture of my journey and the unique value I can bring to projects, visit daveliew.com. Whether you're navigating career transitions or building meaningful things in this AI-driven world, I'd love to exchange ideas and explore how we might create positive impact together.
Comments