Physics in the Age of AI

Opus 4.6 and Codex 5.3 came out yesterday, continuing humanity's march into a glorious future - or our march into irrelevance, depending on who you listen to. Either way, AI is going to take over a lot of skills, maybe even entire jobs, that we previously thought only humans could do.

So what role do skills play in this future?

Some would have you believe that it's just like the gym - now that manual labor is no longer required, we go to the gym for health, happiness, and aesthetics. In their version of the future, doing an hour of math and physics a day is something you do just to keep your own brain healthy or to signal to others that you're intelligent. Like a more interesting version of sudoku or crossword puzzle.

While there are absolutely psychological benefits to learning physics, let's be real - if there aren't any tangible benefits, you might as well go play a challenging videogame.

So to take a glimpse into what's going to happen for engineers, let's look at what's currently happening to programmers.

There was once a time, a short 15 years ago, where you could write a simple CRUD app and immediately be given a job offer close to 6-figures. This was the bottom rung on the ladder. Climb the ladder and you start making more decisions, first about just the code, and then working with the rest of the business to make long-term decisions about the architecture and the customer journey.

CRUD apps can now be trivially one-shot by all frontier models.

The bottom rung is gone.

However, this does not mean that no one will ever get a junior dev job again. It's just that your first job will now be very different.

Certain skills, like being able to write good syntax and debug compiler errors, used to be sufficient to get you a junior dev job. Now LLMs can fill that role, while working at 10x the speed of any human, and they're only going to get better.

Now, junior dev roles are basically engineering manager. Technical skill is important, otherwise your LLMs can make a ton of bad decisions and bullshit you at every turn, but you're not typing the code yourself. The good news is that building the skills required for this can be done 10x faster than in the old days, so you don't have to labor unpaid for 4-5 years before getting this new type of management job.

Now, applying this to engineering, it's even more important that you understand the basics before sending off your robotic minions to work. If you vibe-code without any understanding of how code works, you'll end up with an unmaintainable mess. If you vibe-build a bridge without any understanding of physics, it will collapse and kill hundreds.

You're not going to be doing the calculations yourself (except for learning purposes), but that's already true today. Calculators exist. Numpy exists. CAD exists.

We're just going to go up one level.

Everyone will be a manager. And just like human employees, the LLM gets much better results when their manager has enough technical understanding to guide them.