Insights | NextLink Labs

Most of Your Engineers Should Not Become AI Engineers

Written by Jordan Saunders | Aug 26, 2026, 3:37:43 PM

Every engineering leader is hearing the same advice right now: retrain the whole team as AI engineers before it is too late. That advice is going to waste a lot of money, and I want to walk through why.

The advice conflates two different skills. Using AI to build software faster is one skill. Building software out of AI is a different skill. They sound like the same thing, and they are not close.

Two Tracks, Not One

The power user track — for almost everyone. Your backend developer is still a backend developer. What changes is throughput: agentic coding tools doing the first pass, conventions that tell those tools how your shop works, review habits that catch what the model gets wrong, and the judgment to know which tasks to hand off and which to keep. This is weeks of deliberate practice, not a career change. The engineers on this track keep the craft you already paid a decade for and get faster at it.

 

The specialist track — for a handful. Building production systems where a model is a component means evals, cost behavior, model selection, failure modes, guardrails, and knowing why an agent went sideways at 2am. It is genuine engineering discipline and it is not for everyone, because it does not need to be. In most mid-market organizations, this is a handful of people. Sometimes it is two.

 

Where the Retrain-Everyone Advice Fails

Push your whole team toward the specialist track and most of them will not want it, the organization does not need twenty of them, and you will trade away depth you already own for shallow coverage of a skill the market only needs from a few. A great database engineer who uses AI well is worth more to you than a mediocre AI engineer, and that stays true for as long as I can see.

But the opposite mistake is just as common. Some leaders read the hype fatigue correctly and conclude nobody needs to change. That fails too, because somebody has to own the AI systems your company is about to depend on, and if nobody inside owns them, you will rent that judgment forever.

The Actual Job Is Sorting

The sorting is not complicated. The power user track is for everyone, full stop. The specialist track is for the one or two people who were already experimenting before anyone asked them to. That last part matters more than seniority or title. The engineers who should own your AI systems announce themselves by what they do on a slow Wednesday. You are not assigning the track. You are noticing it.

I run a firm that sells engineering enablement, so weigh my incentive. But notice that the advice cuts against the bigger engagement. The expensive version of this is retraining everyone. The correct version is cheaper.

The Honest Tradeoffs

The power user track is not free just because it is short. Without shared conventions and real review discipline, an AI-accelerated team ships AI-accelerated defects, and you will not see it in the velocity numbers until it is in production. Measure defect rates alongside throughput or you are measuring half the story.

And your specialists need an actual system to build, not a title change. A specialist with no production AI system to own is just an enthusiast with a new business card.

For the Engineers Reading This

The fear underneath all of this is real. The panic version says your craft is obsolete. I think the opposite is happening. When the first draft of everything gets cheap, judgment becomes the bottleneck, and judgment is the part of your craft that took the longest to build. The engineers in trouble are not the ones who skip the AI specialist track. They are the ones who refuse the power user track and keep typing everything by hand out of principle.

The Exercise for This Week

List your engineers. Mark who is using AI tools daily, occasionally, and not at all.

 
The dailies are fine, and one or two of them are probably your specialist candidates already.
 
The occasionals need conventions and a nudge, which is a management problem, not a training problem.
 
The holdouts need a direct conversation about why.

Then answer one question about yourself. Do you know, today, who owns AI systems in your organization? If a name did not come to mind, that is the gap, and it is not solved by sending everyone to a course.