Hire an AI engineer, or bring in a consultant
Three conditions decide it, and headcount is not one of them.
Hire when AI is a permanent part of the product and you can keep someone busy. Bring someone in to build the first system and decide if there should be a second. Most companies need that first.
Is it a fit?
Permanence
AI becomes part of the product or a standing part of operations.
Supervision
Somebody in house can direct the work and judge it.
A second system
There is clearly more than one thing to build.
Start here. Say where the hours go and you get the first place AI would pay.
Where AI pays firstWork it out in a minute
Put in how often it happens and how long it takes. Get hours a week, hours a year, and what it costs a year.
What you get
Hire when it is going into the product
If the model is part of what you sell, it needs somebody whose job is to watch it drift, retrain it and answer for it at two in the morning. That is a role, not a project, and it does not outsource well.
Bring somebody in when nothing exists yet
The first system is the one most likely to be the wrong system. Paying for the learning once, from somebody who has already made those mistakes elsewhere, is cheaper than hiring a person to make them on your payroll.
You cannot supervise what you cannot judge
A first AI hire into a company with no AI experience has no technical manager. They will set their own priorities, mark their own homework, and nobody will know for a year whether the work was good.
Count the weeks of real work, not the headcount
If the right answer is one solid project and then maintenance, that is not a role. A strong engineer with two days of real work a week leaves inside a year, and takes everything they knew about your systems with them.
The hybrid that works
Somebody outside builds the first system and writes down how it runs. Somebody inside, often an existing engineer or analyst, owns it from day one and is trained while it is built. The knowledge lands in the company either way.
The market is a real constraint
AI engineers are expensive, mobile and mostly employed. A small company competing on salary will lose, and will lose again in eighteen months. Design for that rather than assuming you will be the exception.
The question underneath the question
Most companies asking this have not yet decided whether AI is going to be a permanent part of how they work or one good project. That is the actual question, and it is usually unanswerable until one thing has been built and run for a quarter.
The trap on the consulting side is dependency. A system built by somebody who keeps the accounts, the prompts and the documentation is a subscription with extra steps. Ownership and a handover, written down, are what stop that.
Questions we get
How much work justifies a hire?
Roughly, enough that somebody is busy four days a week for a year, and somebody else in the company can tell whether their work is good.
Below either threshold you are creating a role that will be under-used, under-managed, or both.
Can an existing engineer take this on?
Often, and it is usually the best answer available. Somebody who already knows your systems and your data learns the model side faster than an outside expert learns your business.
Give them real time for it rather than adding it to a full week.
What about a fractional AI lead?
It works when the need is judgment rather than hands: what to build, in what order, what not to touch. It works badly when somebody needs to be in the code every day.
Be honest about which of the two you are short of.
Will a consultant leave us dependent?
Only if the contract lets them. Say in writing that the code, prompts, data and accounts are yours, that documentation is a deliverable, and that somebody on your side is trained while it is built.
A firm that resists any of those has told you something useful.
What you are buying comes before who builds it. ChatGPT Enterprise or custom AI. Either way, preparing for AI is what you should have done before they arrive. The same question on the operations side is a fractional COO or a full time COO.
More in the guides and every answer in one place.
Read next
How to judge outside help, the cheaper answers before either, and why most of these stall whoever does them.
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