What is a subject matter expert in the AI era?

Developing subject matter experts is contingent upon putting in the work, getting in the weeds, and learning how to get out of them. For software developers, starting out as an average coder is the first step. But with AI agents increasingly taking on beginner tasks, are we seeing the end of subject matter experts, or is expertise becoming easier to reach? 

AI has created a lot of average software — a task previously part of the learning curve for new developers. And it’s decidedly changed things, creating an entire category of tasks that can be solved by throwing money at them.

In the last two years, we’ve seen AI development move from pilot project to full scale production.

AI makes software fast (not fast software).

Most applications are not cutting-edge and were never meant to be.

By definition, they were built by standard teams, on modest budgets. But they articulate a business process with as little friction as possible, and they’re reasonably reliable given their constraints. People who use those applications don’t really care whether a form submits in 10ms or 50ms, because there’s a threshold past which incremental performance doesn’t have any impact on a user.

AI is very good at clearing that threshold (though maybe we should care more about performance).

AI reduced the barrier to entry, and supply reacted. 

Companies have more software today than at any point in history. This is true on both sides of the transaction, both in the software they purchase from vendors and the software they’re willing to produce themselves.

We value things that solve the problems quickly, and flexible widgets that address multiple problems tend to be more valuable. With AI, everyone is a generalist. You can create marketing copy, NGINX configurations (of some reasonable quality), and back-end code while sipping your morning coffee.

There’s a reason most of us don’t own a grapefruit spoon. It’s a fine tool, but it’s highly specialized, and we’ve got ready alternatives on hand that solve the same problem. Generalists have value.

AI has made it difficult to get a sense of what a developer understands by just looking at their code.

Not long ago, a developer’s utility was tied, fairly directly, to how quickly they could type out the syntax of an application. They would hit the proverbial speed bumps of “syntax,” “architecture patterns,” and “wrangling dependencies,” and their output would reflect that. Today, that constraint is gone. A junior developer’s capacity for generating code exceeds that of the mortals of yesteryear.

So, how should project owners and leadership account for this? How important is developer knowledge, muscle memory, and context in the AI era? And how should businesses build effective software teams composed of subject matter experts within this new dynamic?

Being able to produce average software used to be the gateway to developing subject matter expertise. It was a step along an essential pathway. It was difficult to acquire. It took years of experience to develop. And because it took years, something else happened along the way that we rarely accounted for: people learned the domain. Developers sat in meetings where the requirement was wrong. They shipped the thing that technically worked and watched it fail for a reason nobody had written down. And it was so painful to get it wrong that they fought hard to get it right.

Subject matter expertise was never taught. It is a byproduct of struggling with a problem. That’s the AI gap. If the model does the struggling, the subject matter expertise never forms.

So domain knowledge isn’t being lost because it stopped mattering. It’s at risk because the thing that led to it stopped happening.

The gate moved. It used to sit in front of “Can you build it?” Now it sits in front of “Do you know what to build?” and “What will it cost you?”

There’s a reason a plumber’s billable rate sometimes rivals a lawyer’s. We don’t hire a plumber because we don’t know how pipes work. We hire one because we want the right fix, applied fast, by someone who’s seen what the wrong fix costs. Knowing where and how to apply the fix has always outstripped the difficulty of the task.

The same is true for developers. And that gap just got wider.

The kind of software we’re getting out of the gate with AI is easier to identify. It’s average. The most common tells are that while the software gets the job done, it’s found to be:

At Brightly, our answer is to hire people with the plumber’s expertise who write software that is:

Find where your skill set produces above-average outcomes. How you structure code. How you work with vendors. How you plumb environments together. You’re looking for the work that requires context and experience, and that can’t be replicated by someone with a Claude subscription and a free afternoon.

Code review, copy, and prose, the call on which features get built. Those are the judgment surfaces. If the decisions you’re making are worse than the ones the model would make, you’re in an untenable position. Fix that.

People used to acquire domain knowledge by working slowly and methodically. Almost nobody funds that now. You’ll have to construct the struggle on purpose:

When you do this, two things happen.

Subject matter expertise isn’t gone. It’s concentrated in a cohort of developers who struggled with challenges, tackled missteps, and learned along the way — and who, in turn, are the most valuable engineers on the market. They are also the most equipped to train and steer AI-enabled development projects.

Those organizations that understand that manufacturing ways to develop subject matter expertise is as critical as AI upskilling (deliberately and at real cost) will be the ones who end up holding the only scarce thing left.