The Developer Is Becoming the Reviewer

AI coding assistants arrived.

Now we're entering something different.

AI doesn't just autocomplete the next line anymore. Coding agents can plan tasks, modify multiple files, run tests, inspect failures, and iterate. A recent JetBrains survey found that 90% of professional developers surveyed were already using AI coding agents at least weekly.

So the interesting question isn't:

It already does.

The more interesting question is:

Software engineering has always been more than typing.

We design architectures.

We make trade-offs.

We decide what should be built.

We investigate bugs.

We review code.

We think about security and performance.

AI is increasingly taking over one of the most visible parts of the job: implementation.

That doesn't necessarily make engineers less important.

It changes where their attention goes.

The developer may increasingly become the person who defines the problem, gives the agent enough context to solve it, evaluates the solution, and decides whether it should actually reach production.

In other words:

There's a catch.

If an AI agent writes a thousand lines of code, you still own those thousand lines.

If the code contains a security vulnerability, the AI won't be the one explaining it during an incident.

If the architecture is wrong, generating the code faster only gets you to the wrong architecture faster.

And if you don't understand what the agent produced, "AI-assisted development" can very quickly become "AI-dependent development."

That's why I don't think the answer is to stop learning programming.

I think the opposite is true.

When implementation becomes cheaper, understanding becomes more valuable.

You need to know enough about algorithms to recognize an inefficient solution.

Enough about databases to question a bad query.

Enough about networking to understand why a distributed system is failing.

Enough about security to spot a dangerous shortcut.

Enough about software architecture to know when an agent has solved the wrong problem beautifully.

The irony is that AI may make fundamental computer science knowledge more important, not less.

There's also an important distinction between using an AI assistant and delegating an entire engineering task to an agent.

Developers still overwhelmingly keep humans in the loop. JetBrains' 2026 research found that, on average, developers reported about 10% of their code as fully agent-generated, compared with roughly 25% written with AI assistance and 75% manually written.

So the future probably isn't:

It's closer to:

And that workflow is still evolving.

As a computer science student, this changes what I want to become good at.

Of course I want to write good code.

But I also want to become good at something harder:

I want to be able to enter an unfamiliar codebase, understand the architecture, define a problem precisely, give an AI agent useful constraints, inspect what it produces, and know when its solution is wrong.

Because if AI is becoming better at writing software, perhaps the competitive advantage isn't being the fastest person at writing code.

Maybe it's being the person who can direct the fastest system effectively.

The developer isn't disappearing.

The job is moving up a level.

And I think we're only beginning to see what that means.

#AI #SoftwareEngineering #CodingAgents #ArtificialIntelligence #ComputerScience #Tech #FutureOfWork