Hey,
Over the last three newsletters, I looked at three different parts of ChatGPT's integration model.
Apps connect ChatGPT to external systems such as Gmail, Drive, Calendar, Slack or Notion.
Plugins package reusable capabilities and can bring different pieces together.
Skills capture a repeatable method for how you want a task handled.
Separately, those ideas are fairly simple.
The useful part is understanding what happens when they work together.
One request can use all three
Suppose you ask ChatGPT:
I have a meeting with ABC Ltd tomorrow. Prepare me for it. Tell me the latest status, what we have promised, what they still owe us, what we still owe them and the five points I should cover on the call.
That looks like one request.
But underneath it, ChatGPT may need several different capabilities.
The important thing is that each piece has a different job.
Apps provide the information
The apps answer:
Where does ChatGPT need to look?
For this meeting, the information could be spread across several systems:
Calendar for the meeting details;
Gmail for recent correspondence;
Drive for the latest proposal or scope;
a CRM or Notion database for current status and notes.
Without those connections, you would usually have to find the information yourself and paste it into ChatGPT.
With the relevant apps connected, ChatGPT can retrieve supported information directly from those systems, within the permissions of the connected account.
That is the role of the app layer.
The skill provides the method
Finding the information is only part of the job.
ChatGPT still needs to know what to do with it.
A meeting-preparation skill could contain instructions such as:
identify the latest known status;
separate our commitments from the client's commitments;
identify anything still outstanding;
find unresolved questions;
highlight anything that appears inconsistent;
produce a short meeting brief;
do not invent missing information.
That is the method.
The skill does not provide access to Gmail or Drive. It defines how the task should be handled once the relevant information is available.
The plugin packages the capability
Now put those pieces together.
The user does not really care about "using Gmail", "using Drive" or "running a meeting-preparation skill".
The actual job is:
Prepare me for this client meeting.
A plugin can package that capability so the user does not have to think about each component separately every time.
That is the distinction I find most useful:
Apps give access.
Skills provide the method.
Plugins package the capability.
Permissions still sit around the whole workflow
There is one more piece that is easy to overlook: permissions.
Connecting an app does not give ChatGPT unrestricted access to that system.
The connected account still determines what information is available. The app itself determines which operations are supported. Some write actions may also require approval, and workspace administrators can restrict which apps or plugins are available.
So in the meeting example, ChatGPT might be able to read recent correspondence and prepare a follow-up email without necessarily being allowed to send it automatically.
The workflow still operates inside those boundaries.
What changes compared with normal ChatGPT use?
The difference is mostly about who gathers the context and how repeatable the process becomes.
Without integrations:
Find the email
→ find the proposal
→ check the calendar
→ copy everything into ChatGPT
→ explain how you want it analysed
→ get the meeting brief
With the pieces connected:
Ask for the meeting brief
→ apps retrieve the relevant context
→ the skill applies the method
→ ChatGPT produces the result
The reasoning task may be similar.
What changes is that you spend less time moving information between systems and re-explaining the process.
Start with the task, not the integrations
After looking at apps, plugins and skills separately, this is probably the main point I would take from the series.
Do not start with:
What can I connect to ChatGPT?
Start with a repeatable piece of work.
Then ask four questions:
What is the task? What are you actually trying to get done?
What information does it need? Which systems contain that information?
What method should be followed? Is there a repeatable way you want the work handled?
What should ChatGPT be allowed to do? Read information only, prepare something for review, or carry out a supported action?
Those four questions are usually enough to tell you whether an app, skill, plugin or some combination of them is useful.
The whole series in one sentence
If I had to reduce the four newsletters to one model, it would be this:
Apps connect ChatGPT to the information and actions it needs. Skills define how the work should be done. Plugins package those pieces into reusable capabilities. Permissions decide what can actually happen.
That is the model I would use when looking at any ChatGPT integration.
Instead of asking how many systems you can connect, pick one task that currently involves moving information between tools and ask whether these pieces can make that task simpler.
That is where the integrations start becoming useful.