LJL learning experience

Skills or GPTs?

Determine when it is appropriate to use a Skill versus a GPT by learning the distinction, testing the decision lens, and working your way through a branching design challenge.

The heartbeat of a Skill A reusable way of completing work that should travel across changing inputs.
The heartbeat of a GPT A configured assistant experience people return to for a defined kind of help.
orientation

Begin beneath the name.

Purpose first · Tool second

Imagine that someone brings you a new idea for using ChatGPT. They want people to stop rebuilding the same prompts. They want the work to feel more consistent. They want ChatGPT to understand how their team operates.

Then comes the question: should this become a Skill or a GPT?

At first, the answer may feel less obvious than it should. Skills and GPTs can both contain instructions. Both can shape how ChatGPT responds. Both can support work that happens more than once. That overlap is real.

But the decision is not arbitrary. The clearest move is to stop asking, Which tool could do this? and start asking, What does this solution need to preserve?

fully teach it

Meet the Skill and meet the GPT.

Both can be useful. They are not interchangeable. One protects a method. The other protects an experience.

meet the skill

The method travels.

A Skill is a reusable workflow that helps ChatGPT perform a particular kind of task consistently. It can include instructions, examples, supporting resources, and code. What makes it valuable is not one document, one conversation, or one topic. What makes it valuable is the stable way of working.

same sequence same checks same output structure
Input 1
Customer interview set
Input 2
Different customer story
Input 3
New project notes
→
Stable workflow
Identify the central need.
Separate evidence from interpretation.
Group themes and flag gaps.
Organize the final brief in the same structure.
→
Consistent result
Insight brief with the same quality checks every time.

Think about the customer-interview example. The people change. The questions change. The stories change. But every time the work is complete, you want ChatGPT to follow the same careful method. The interviews are not reusable. The method is.

meet the gpt

The experience stays recognizable.

A GPT is a version of ChatGPT configured for a particular purpose. It can combine instructions, knowledge, conversation starters, and capabilities to create a tailored assistant experience. What makes it valuable is the stable purpose people return to when their questions vary.

stable role selected knowledge clear boundaries
What does this term mean?
Where do I find this resource?
How do these departments connect?
Onboarding GPT
trusted support home
Help me prepare for a manager conversation.
What is the policy behind this process?
How should I start this task?

Think about the new-employee example. The questions move in different directions. The need is not one shared procedure. The need is a recognizable assistant with a stable purpose, a trusted body of knowledge, and boundaries around the support it provides.

why it matters

When the structure does not match the work, problems begin quietly.

You can force either tool to do many things. That does not mean it is the strongest design. Misalignment usually shows up as drift, duplication, or blurry boundaries.

Overbuilding the experience

Sometimes a team builds an entire GPT for one narrow review process. It may work. But the work may only need a dependable method, not a full assistant experience.

One review function ≠ Whole assistant world

Blurring the workflow

Sometimes a team keeps adding unrelated work into one Skill. It begins to do coaching, policy interpretation, planning, and review all at once. The more it carries, the harder its method is to recognize.

review coaching policy planning

Creating two versions of the truth

Sometimes a team creates both a Skill and a GPT for the same job. At first it feels complete. Then one is updated and the other is forgotten. Now there are two places to maintain the same process.

Skill updated ↔ GPT not updated

The same topic can need different tools.

The subject matter does not determine the tool. Two solutions can live in the same domain and still need completely different structures.

same topic · different structure

When the need points toward a Skill

Imagine that a learning team receives workshop notes from different subject matter experts. One workshop is about leadership. Another is about clinical communication. A third is about project management. The content changes every time.

But the team always needs ChatGPT to begin in the same place. It must identify who the learners are and what they need. It must clarify the learning goal. It must separate essential teaching from information that is merely interesting. It must shape the material into the approved lesson structure and identify what is missing before the draft moves forward.

Pause here.
The team is not trying to preserve one topic, one set of notes, or one conversation. The team is trying to preserve its way of developing a lesson. The materials change. The method should remain stable. That points toward a Skill.

When the need points toward a GPT

Now imagine that the same team wants to create a Learning Design Coach. One designer wants help strengthening an objective. Another wants help choosing an interaction. Someone else wants an explanation of cognitive load. Another person wants two possible approaches compared.

There is no single sequence that every conversation will follow. What should remain stable is the coach’s purpose, its knowledge, the perspective it brings, the boundaries around its advice, and the kind of support people expect when they return.

Pause here.
The topic is still learning design. That did not determine the tool. The structure of the need did. What needs a home now is a recognizable support experience. That points toward a GPT.
when both are appropriate

Both can be right when the jobs are different.

The presence of two tools is not the problem. Duplication is the problem. A Skill and a GPT can work together when each one protects a different source of value.

Separate the method from the experience.

Imagine that employees need help understanding organizational policies. Their questions may vary, but they need one recognizable place to ask for support grounded in approved information. That is an assistant experience. A GPT can give it a home.

Now imagine those same employees must complete incident reports. Every report must pass through the same checks. Required details must be present. Missing evidence must be flagged. The final review must follow a consistent structure before submission. That is a method. A Skill can preserve it.

Using both makes sense because the jobs are different. One supports varied policy conversations. The other protects a repeatable review process.

One ecosystem. Two distinct jobs.

Policy Support GPT

Different questions enter a stable support experience with selected knowledge, a clear role, and clear boundaries.

+
Incident Review Skill

Different reports move through the same checks, required fields, evidence review, and output structure.

The name does not decide.
Frequency does not decide.
Complexity does not decide.
The ability to hold instructions does not decide.
a simple analogy

Think recipe and restaurant.

A strong analogy can make a slippery distinction much easier to hold onto.

Skill

A recipe carries a method.

It tells you what you need, what to do, what order matters, what to watch for, and what the finished result should become. You can use the recipe in a different kitchen. You can use different ingredients. The method still travels.

Ingredients may changedifferent inputs
Order still mattersstable sequence
Checks still matterquality criteria
Result has a shapeconsistent output
GPT

A restaurant carries an experience.

It has a purpose. It has an identity. It has resources, boundaries, and a recognizable type of experience. People choose it because of the kind of support it provides. A restaurant may depend on many repeatable methods. But a recipe does not need an entire restaurant built around it.

Reason people returnstable purpose
What it knowsselected knowledge
How it helpsconsistent support
What it will not doclear boundaries
the decision lens

Ask three questions and watch the answer sharpen.

When you are uncertain, do not start with the label. Start by looking for the part that must remain steady.

Question 1 Question one

What are you really trying to preserve?

Question 2 Question two

What will change the next time this is used?

Question 3 Question three

Where does the main value live?

Now step into the branching simulation.

You are the advisor. Every decision moves the architecture forward. You will see what your recommendation solves, what it leaves exposed, and what happens when the need changes.

chapter 1

The First Request

final challenge

Look beneath the title one more time.

The name may sound like a dedicated assistant. The evidence tells a more important story.

The Launch Readiness Advisor

Read the case carefully.

A product team wants to create a resource called the Launch Readiness Advisor. The name makes it sound like a dedicated assistant. Now look beneath the name.

What changes Different launch plans arrive for different products.
What stays steady Every plan must be checked against the same ten readiness criteria.
What stays steady Missing evidence must be identified and each area must be rated using the same scale.
What stays steady Findings must be organized into the same readiness-report structure and human judgment must be flagged.
What is not needed Users are not expected to explore broader launch questions or seek ongoing coaching.
your call

What is the stronger starting point?

Choose the option you believe best matches the work. Then check your reasoning.
reflection pause

Make the need visible in your own work.

Picture someone using a solution you are considering. Then picture a different person using it a month later. What changes? What must remain recognizable?

Your reflection saves in this browser so you can return to it.