Build
Prompting Best Practices
Best practices for writing effective prompts that make your Pickaxes focused and reliable.
The instructions you give your Pickaxe are the single biggest factor in how well it performs. A well-written prompt turns a generic chatbot into a focused, reliable tool. Here are the best practices for getting the most out of your agents.
Be specific about the role
Start by telling your Pickaxe exactly who it is and what it does. A clear role definition helps the AI stay focused and avoid going off-topic.
Weak: "You are a helpful assistant."
Strong: "You are a customer support agent for a SaaS product called Acme. You help users troubleshoot account issues, answer billing questions, and guide them through common workflows. You do not provide legal or financial advice."
Define boundaries
Tell your Pickaxe what it should and shouldn't do. Setting clear boundaries prevents the AI from overstepping or giving responses you don't want.
- What topics are off-limits?
- Should it ever say "I don't know"?
- Should it escalate certain questions to a human?
- Are there any actions it should never take on its own?
Set the tone and style
If tone matters for your use case, spell it out. The AI will match whatever style you describe.
- "Respond in a friendly, casual tone."
- "Keep responses professional and concise — no more than 2-3 sentences."
- "Use simple language that a non-technical user can understand."
Structure your instructions
Long, unstructured prompts are harder for the AI to follow. Break your instructions into clear sections using headers, bullet points, or numbered lists.
## Your role
You are a fitness coach that helps users build workout plans.
## Rules
- Always ask about the user's fitness level before recommending exercises.
- Never recommend exercises that require equipment unless the user confirms they have access.
- Keep responses encouraging and supportive.
## Response format
- Start with a brief explanation of the workout goal.
- List exercises with sets and reps.
- End with a motivational tip.
Use the model reminder
The model reminder is prepended to every user message and is a great place for instructions the AI should never forget. Use it for your most critical rules — things like:
- "Never share pricing information — always direct users to the pricing page."
- "Always respond in the same language the user writes in."
- "If the user asks about cancellation, offer to connect them with support first."
Give examples
If you want the AI to respond in a specific format or follow a particular pattern, include an example directly in your prompt. Examples are one of the most effective ways to guide behavior.
When a user asks for a recipe, respond in this format:
**[Recipe Name]**
Prep time: [X minutes]
Serves: [X]
Ingredients:
- [ingredient 1]
- [ingredient 2]
Steps:
1. [step 1]
2. [step 2]
Keep actions in mind
If your Pickaxe has actions connected, make sure your prompt tells the AI when to use them. Be explicit about the trigger conditions — both in the main role prompt and in each action's trigger prompt.
Example: "When a user asks to schedule a meeting, use the calendar booking action. Do not try to schedule anything through the conversation itself."
Test and iterate
Building a great Pickaxe is an iterative process. Use the Preview tab to test your agent and refine your prompts based on how it responds.
- Try edge cases — what happens when a user asks something unexpected?
- Test with different user personas to see if the tone and responses hold up.
- Use the AI Helper to quickly experiment with changes and see their impact.
If your Pickaxe gives a bad response, don't just add more instructions — first figure out why it happened. Sometimes the fix is removing conflicting rules rather than adding new ones.
Quick reference
| Tip | Why it matters |
|---|---|
| Define a clear role | Keeps the AI focused and on-topic |
| Set boundaries | Prevents unwanted responses |
| Specify tone and style | Ensures a consistent user experience |
| Structure your prompt | Makes complex instructions easier to follow |
| Use the model reminder | Reinforces critical rules on every message |
| Give examples | Shows the AI exactly what you expect |
| Be explicit about actions | Helps the AI trigger the right action at the right time |
| Test and iterate | Catches issues before your users do |
