The Most Expensive Mistake in AI Hiring: More AI Engineers

Nine months ago, a healthcare client asked us to help them build an AI workflow for claims processing. They had already hired two machine learning engineers. Both were talented. Both built a technically elegant system that processed claims faster than anything the team had seen before.

It also misclassified 23% of claims in the first week. Not because the model was wrong. Because neither engineer understood that certain procedure codes in that client’s network map to different reimbursement tiers depending on the provider’s contract type. That is not in any training data. It lives in the head of the senior claims reviewer who has been doing this for fourteen years.

We rebuilt the workflow around her. She is not an engineer. She does not know what a large language model is. She is now the most productive person on that team by a wide margin, because she knows what “correct” looks like for every edge case the model has never seen. The engineers built the system. She taught it how the business actually works.

AI Can Do the Specialist Work. It Cannot Do the Connecting.

Here is what changed and most hiring managers have not caught up to it yet. AI is now good enough to do the narrow, repetitive, skilled work that used to require a dedicated specialist. It can write code, analyse data, draft legal memos, format reports, process documents, run calculations. Not perfectly. But well enough that the specialist’s advantage over a capable generalist has shrunk to almost nothing on these tasks.

What AI cannot do is connect the output to the business. It cannot look at a claims decision and know that this particular provider has a history of billing disputes. It cannot review a financial report and notice that the numbers are technically correct but tell the wrong story to the board. It cannot read a product spec and feel that the feature makes sense technically but will confuse the customer segment it is designed for.

That connecting work, the judgement that sits between the output and the decision, is what generalists do. It is also what most companies are not hiring for, because the job title does not exist in their HR system.

“You can’t hire someone who has been building AI agents for five years. The technology hasn’t existed that long. The people thriving are the ones who learn fast, adapt fast, and act without waiting for direction.”
That observation from a VentureBeat analysis matches what we see in every engagement. The best people on AI teams are not the ones with the most AI knowledge. They are the ones with the most business knowledge who figured out how to use the tools.

The Two People Who Stall Every AI Project (And the One Who Saves It)

We walk into this pattern so often we have a name for it internally. Two profiles sit on opposite sides of a conference table. Neither is wrong. Both are stuck.

Person one: the engineer. Technically brilliant. Built the pipeline, tuned the model, optimised the prompts. Has no interest in learning why the finance team runs month end close the way they do, or why the compliance team rejects certain claim types on Thursdays. Thinks the business should adapt to what the system can do. Builds tools that are impressive and unused.

Person two: the business operator. Knows every exception, every workaround, every reason why the current process exists. Has been told AI will “help” but nobody showed them how. Uses AI the way they use Google: types a question, gets a paragraph, skims it, goes back to the old way. Decided months ago that “it doesn’t really work for what we do.”

Person three: the one you need. Understands the business deeply enough to know what “right” looks like. Understands the technology enough to know what is possible. Sits between the two sides and translates. Briefs the AI the way you would brief a sharp new hire: here is who we are, here is the project, here is what good looks like, here is what to avoid. Anthropic published research last week showing that this kind of user, the one who gives better context, iterates, and pushes back on bad output, is pulling measurably ahead of everyone else. The gap is growing every month.

That third person is a generalist. They might have come from operations, from product, from client services. They probably do not have “AI” anywhere in their job title. But they are the reason the project produces results instead of a demo.

You Already Have This Person. Here Is How to Spot Them.

Every company we work with already has this person somewhere. They are never on the AI team. They are usually in operations, finance, compliance, or client services. They have been doing their job long enough that they carry the institutional knowledge nobody wrote down. Here is what they look like when you put AI tools in front of them.

1. In the first week, they do not ask how the tool works. They ask what it can read. They want to know: can I feed it our reconciliation rules? Can it see the contract terms? Can I point it at our claims history? The specialist starts with the model. The generalist starts with the data that matters to the business.
2. By day ten, they have built something nobody asked for. Not a prototype. Not a proof of concept. A working shortcut for something that used to take their team half a day. It might be ugly. It probably violates every best practice the engineering team would insist on. But it works, because it was designed by someone who knows exactly what “works” means for that specific task.
3. They catch errors the model cannot catch. This is the part that surprises people. The AI produces output that looks polished and plausible. The generalist reads it and says “that’s wrong” in five seconds, because they know that this vendor always double bills on multi currency invoices, or that this regulatory clause changed in the last update, or that this customer segment behaves differently than the data suggests. That judgement is not trainable. It is earned.
4. They become the translator nobody hired. When the engineering team builds something, this person is the first to say “that is not how we actually do it” and the only one who can explain why in terms the engineers understand. When the business team says “AI does not work for what we do,” this person is the one who quietly proves them wrong by producing results with the same tools everyone else gave up on.

You probably already know who this is in your company. The person the team goes to when something does not make sense. The one with the institutional memory. The one who has been doing the job so long that they are the documentation. That is your most important AI hire, and they are already on your payroll.

What Your Next Three Hires Should Look Like

If your current AI investment is underperforming, and most are, the instinct is to hire more AI specialists. Resist it. You probably have enough technical talent. What you are missing is someone who can connect what the technology does to what the business needs.

Here is what we tell clients when they ask us who to hire next:

1. Hire for business depth first. The person who has spent a decade in your industry and your function is worth more in an AI team than a freshly credentialed ML engineer. They know the exceptions, the edge cases, the reasons why things are done a certain way. That knowledge is the context AI needs to produce useful output. Without it, you get technically correct results that miss the point.
2. Hire for curiosity second. The right person does not need to know how a model works. They need to be the kind of person who, when you hand them a new tool, will spend a weekend figuring out how to make it do something nobody intended. Researchers at Stanford and Harvard found that AI equalises performance across specialties. Give a marketing generalist AI tools and they perform on par with a dedicated web analyst. The domain knowledge carries. The narrow specialism does not.
3. Hire for translation third. The ability to sit between the technical team and the business team and make each side understand the other is the scarcest skill in enterprise AI right now. It is also the one nobody puts in a job description, because HR does not have a template for it.

Depth Is Cheap Now. Width Is Expensive. Staff Accordingly.

For twenty years, the premium went to depth. The person who knew one thing cold was the most valuable hire. AI rewired that equation. The narrow skill that took a decade to build can now be approximated by a machine in seconds. It is not perfect, but it is good enough to shift the economics.

The premium now goes to width. The person who can move across domains, who can brief AI with the context it lacks, who can look at a technically correct output and say “that is not what we need” is worth more than they have ever been. They are also harder to find, because nobody trained them, nobody gave them a title, and most of them do not know they are exactly what the market is looking for.

Go find them before your competitor does. They are already in your company. They are the ones who know why everything works the way it works. Give them the tools and get out of their way.

OP helps companies figure out who should be on their AI teams, how those teams should be structured, and what the workflows should look like. If you have the tools but not the results, the problem is probably not the technology. Let’s talk about the people side.