← Mission log index
LOG-003

Start New UI Design with Claude Design System Prompt

AI HELPS WITH UI DESIGN · 25 May 2026 · 12 min · RX ONLY · NO COMMENTS

Start new UI design with Claude design system prompt

When I decided to redesign jcus.link, I started with the Claude Design System Prompt created by @Pliny The Liberator. What began as an experiment in AI-assisted design quickly evolved into a complete redesign, resulting in the cleaner, more polished user experience you see on the site today.

There is what I learnt

If your goal is to teach developers how to write effective system prompts, using a real-world prompt as a case study, I would write it as an engineering article rather than a prompt analysis. The focus should be on why the prompt works, not just what it says.


Learning Prompt Engineering from a Real System Prompt

Large Language Models are often described as "the new operating system." If that is true, then prompts are the configuration files that define how the system behaves.

Many developers begin prompt engineering by writing simple instructions:

text
You are a helpful assistant.Answer questions clearly.

While this works for basic use cases, production AI systems require much more structure. As applications become more complex, prompts start to resemble software architecture documents rather than simple instructions.

Recently, I studied a large production-grade system prompt. What caught my attention was not the length of the prompt, but how intentionally it was organized.

The prompt reveals several important principles that every developer can apply when building AI applications.

Principle 1: Define Identity Before Behavior

Most inexperienced prompt writers immediately start describing tasks:

text
Answer questions.Write code.Help users.

Experienced prompt writers start somewhere else:

text
You are an experienced software architect.You prioritize maintainability.You explain tradeoffs clearly.

This establishes identity before behavior.

Humans naturally make decisions based on their roles and responsibilities. Large Language Models behave similarly. When the model understands who it is supposed to be, it can make better decisions in situations not explicitly covered by instructions.

Think of it this way:

text
IdentityDecision MakingBehavior

Without a clear identity, the model has no framework for handling ambiguity.


Principle 2: Separate Responsibilities into Layers

One common mistake is putting everything into a single block of instructions.

For example:

text
Be helpful.Answer questions.Write code.Use markdown.Be friendly.Avoid mistakes.

As prompts grow, this quickly becomes difficult to maintain.

The system prompt I studied was divided into multiple layers:

text
Identity LayerBehavior LayerTool Usage LayerSafety LayerFormatting Layer

This mirrors how software engineers design applications.

Instead of:

text
One giant function

We create:

text
Small focused modules

The same principle applies to prompt engineering.


Principle 3: Use Rules, Not Suggestions

Weak prompts often contain language like:

text
Try to be concise.Usually explain things clearly.Consider providing examples.

These are suggestions.

Strong prompts use explicit rules:

text
Always explain technical decisions.Never expose internal reasoning.Use examples when teaching concepts.

Models follow clear constraints more consistently than vague guidance.

When designing prompts, imagine writing requirements for another engineer. Ambiguous requirements produce ambiguous results.


Principle 4: Handle Edge Cases Explicitly

One of the most impressive aspects of production prompts is how much attention is given to exceptions.

Instead of only saying:

text
Answer user questions.

The prompt defines behavior for special situations:

text
If the user asks for medical advice...If the user uploads a file...If the user requests code...If the user asks for personal information...

This is exactly how software engineers design robust systems.

We do not write:

java
if (happyPath)

and ignore everything else.

We write:

java
if (happyPath)else if (...)else if (...)else

Prompt engineering follows the same philosophy.


Principle 5: Specify Output Formats

One lesson many developers learn quickly:

The model cannot read your mind.

If you want a specific output format, define it.

Poor instruction:

text
Explain the solution.

Better instruction:

text
Provide:1. Problem Summary2. Root Cause3. Recommended Solution4. Example

The system prompt repeatedly defines output structures.

This dramatically improves consistency.

Think of it as designing an API contract.

The more precise the contract, the more predictable the result.


Principle 6: Teach the Model How to Think

The most advanced prompts do not simply tell the model what to do.

They explain how to make decisions.

For example:

text
When multiple solutions exist:- Prefer simplicity- Prefer maintainability- Explain tradeoffs

This transforms the prompt from a task list into a decision framework.

The model becomes more adaptive because it understands priorities rather than memorizing responses.


Principle 7: Prompts Are Software

Many developers still view prompts as text.

I no longer do.

A production prompt is closer to source code than documentation.

It contains:

  • Requirements
  • Constraints
  • Decision trees
  • Formatting rules
  • Error handling
  • Behavioral policies

In traditional software engineering we write:

text
Code controls execution.

In AI engineering:

text
Prompts control execution.

The prompt becomes part of the application architecture.


Final Thoughts

The biggest lesson I learned from studying a production system prompt is that effective prompt engineering is not about finding magical words.

It is about applying the same engineering principles we already use when building software:

  • Define clear responsibilities.
  • Establish decision-making frameworks.
  • Handle edge cases.
  • Create predictable interfaces.
  • Continuously refine behavior.

As AI applications become more sophisticated, prompts will increasingly resemble software design documents rather than casual instructions.

The developers who succeed in the AI era will not be those who write the longest prompts. They will be the ones who design prompts with the same discipline, structure, and intentionality that they bring to software architecture.