ChatGPT Enterprise, or custom AI
Three things decide it, and the feature list is not one of them.
Buy seats for work a person does at a keyboard and checks. Build where work must run on its own, reach your systems, and be checked after. Most companies need both, seats first.
Is it a fit?
Who starts it
A person opening a tab, or something arriving on its own.
What it reaches
The public internet, or records only you hold.
Who checks it
Whoever asked, or a named owner with a route for exceptions.
Start here. Say where the week goes. The tool names where AI would pay first.
Where AI pays firstWork it out in a minute
One task, four questions. Says which shape of AI fits, or that none of them do yet.
What you get
Who starts the work
A seat license waits for a person to open a tab. A built system runs because something arrived. That difference decides more than any feature comparison.
What it has to reach
A general tool knows the public internet and whatever somebody pastes in. It does not know your open orders, your ledger or your queue, and pasting those in is what a policy should prevent.
Who checks the output
With seats the checker is obvious: whoever asked. With a built system checking has to be designed, with a named owner and a route for the ones it gets wrong.
Seats first, almost always
A month of real use produces a list of what people pushed into it and where it stopped being enough. That list is the build brief, and collecting it costs a month rather than a project.
Build one task, not a platform
The companies that get value build a single task end to end and measure it. The ones that do not start with a platform nobody has found a use for yet.
Ownership sits differently
A subscription ends and takes its history with it. Something built stays in your accounts, with the prompts, the logic and the data in your name.
How to decide, in order
Start with who starts the work. If a person opens a tab, describes what they want and checks the answer before using it, that is a seat license, and it is the cheapest useful AI a company can buy.
Then ask who checks. With seats it is whoever asked. With something built, checking has to be designed in, because by definition the output goes somewhere without a person in the middle.
Questions we get
Can we do both?
Almost always, and in that order. Seats cover the desk work. A built system covers what has to run on its own.
The seats also tell you what to build, which is why they come first.
Is a custom GPT the same as building?
No. A custom GPT is a configured seat: it still waits for a person and still only knows what it is handed.
It is a useful middle step, and it is not a system that runs without anybody there.
What about our own data?
That is the question seats cannot answer. Either the system reaches your records properly, or somebody pastes them in, and pasting is what a policy should prevent.
The AI policy page has the wording for that.
How long does a build take?
One task, end to end, is usually weeks rather than quarters, when the data path already exists and somebody owns the output.
Where those are missing, fixing them is the project and the model is not.
More in the guides and every answer in one place.
Read next
How to test the claim before buying either, the rules to write first, and what a built system looks like when it ships.
Shaheer leads the work, with engineers, writers, filers and analysts behind him. C-suite operations for a San Francisco AI company, Six Sigma on the process side, Anthropic certified on the Model Context Protocol, ten years across eight industries. See what we have built