A few years ago, I felt like software engineering was getting boring. All the easy opportunities have been taken, and there was nothing left to do in the industry but make incremental improvements.

Then, LLMs changed everything. Suddenly, there was excitement. Transformation! The opportunities were (and are) endless.

2026H1: Whiplash and a Bitter Lesson

I started the year thinking we were going to work on AI tools like MCP servers. Then, we suddenly pivoted to skills. After we started generating skills, the next generation of models could accomplish the same task without a skill! We learned a bitter lesson.

The churn didn’t end there. Over the course of several months, we tried out different IDEs and CLIs. I discovered a new experimental IDE, loved it, and it was shut down only a week later.

In short, we endured a lot of whiplash in our tools, but the churn seems to be settling down now.

2026H2: Velocity at all Costs

The second half of the year saw an enormous shift. By August, the models and harnesses available to us were a world better than they were a year ago.

Simple tasks that agents would previously stumble with are now routine. The latest agents and harnesses can generate dashboards and beautiful UIs and logging corrections and data analysis docs. Mobile developers are writing changelists for server code. We ship more code and we ship it faster. We are accomplishing more than ever before.

Some engineers push back. But they are overwhelmed. AI-generated code might not be as maintainable and bug-free as artisanal code, but it gets the job done. Besides, we can just have an agent fix the code later.

It’s an exciting time. But it can also feel uncertain. The team has boarded this train, but no one knows where the train is going.

What does this mean in the short term?

I see the team struggling to keep up with the pace of code reviews. We use AI to assist, but tired, overwhelmed engineers stamp changelists and move on. Senior engineers feel this pain the most, since they own more code and complex systems.

I see a lot of AI-generated design docs. Agents provide more low-value information than necessary and make it harder to see the real design decisions. The signal to noise ratio is getting lower. And it’s hard to prove that the author actually thought about the design (which is where human judgement is most needed!).

And people who are not software engineers are writing code (or rather, they’re prompting agents to generate code). PMs, PgMs, UXers… anyone can send a changelist. However, the author might not know why the code works, or if there are subtle bugs lurking in the shadows.

I used to say “Software engineering is an art practiced by few, and understood by fewer.” This is more true today than ever before.

What does this mean in the long term?

No one knows. The practice of software engineering is fundamentally changing.

Engineers already spend more time reviewing code than writing it.

Design and architecture will become more important than ever.

I think the industry will need fewer junior software engineers to create the same output. Layoffs are likely to continue, though they will be uneven through roles and levels.

I don’t think AI is going to lead to a widespread economic collapse. LLMs might be as disruptive as the printing press and movable type. The change will probably be less bad than the worst-case scenario, less positive than the best-case scenario, and different than what most people imagine. (Anyone who generates hype or FOMO via Instagram reels can safely be ignored.)

In any case, adoption moves at human speed.

Where do we go from here?

In order to keep pace with the latest technology, I recommend trial and error. Do some safe experiments and experience the jagged edge for yourself. A course can help but will likely be obsolete in 6 months.

If you study material, do it right. Don’t pretend like you’re learning by scrolling through social media. If you passively watch or listen without taking notes, you’re probably not engaging the part of your brain that retains information.

On a more philosophical note, I wish I had some sage career advice for CS students.

No one knows what’s going to happen next. Everyone, even career professionals, are still figuring out this whole AI thing. Anyone who says they know exactly what’s going on doesn’t know what they’re talking about.

Something weird is happening in software engineering. We’re going to figure it out together.

…probably?