What AI does, and what it doesn’t do

I want AI to program an entire REST API for me from scratch. I want it to create a game, end to end.

AI cannot do these sorts of things—at least not well. Whether it will be able to do them in the future, I’m not sure. But I firmly believe that some level of technical understanding will still be required.

AI creates another layer of abstraction. We don’t program in Assembly anymore; we use high-level languages like Java, C++, Rust, and Go. There’s a lot happening under the hood of these programs. AI simply creates another abstraction on top of these, and that’s where its power lies.

AI isn’t that big of a leap

I view AI simply as a “Google moment”—a foundationally new way to index, search, and apply information.

There were search engines before Google, but Google cornered the market on indexing and searching information effectively. Before Google, it wasn’t uncommon to comb through multiple engines to find what you were looking for. Even with Google, searching for “adjacent information”—pieces of information that are tangentially related to what you’re actually looking for—was still a human-driven effort. You had to gather bits from different sources, synthesize them into a digestible format, and then apply them to your problem.

That’s what AI is breaking down.

Take a common workflow for an engineer—or a bog-standard Linux nerd like myself. The Linux universe is vast. I like Arch (btw), but understanding its nuances—like what package managers are available or the standard tooling for Bluetooth and device drivers—is a broad subject that could take days or even weeks to research.

But what if I just want to get started and actually work on something?

Shortening the feedback cycle

The best teacher is failure.

There’s a great scene at the end of Finding Forrester where the protagonist is being tried for plagiarism, and the author explains that he often starts writing from another author’s work and then sees where he ends up—usually somewhere completely different.

The learning feedback cycle is very real. Trying to solve problems in a vacuum without a place to start or some form of context is a fool’s errand. Animal trails in the woods often become hunting trails for humans. The foundation is laid, and others build upon it.

Now, imagine having a tutor that never gets tired, is right 90% of the time, and is always available.

That’s what AI offers.

The third rail: Environmentalism and cost

AI is not friendly to the environment. Right now, all AI companies are losing money.

If you thought crypto sucked up a ton of power, wait until you see the energy consumption of AI server farms.

Energy is the lifeblood of GDP, and whoever figures out how to make the cheapest, most plentiful energy will be the next superpower of the 21st century. Look closely at France, Norway, and Iceland—all have an abundance of cheap, renewable energy from a mix of nuclear, hydroelectric, and geothermal sources. The U.S. has wind and solar, with hopefully more nuclear coming in the future. It’s simply a matter of building critical mass.

But in the meantime, AI will remain an expensive and resource-intensive tool. That’s not stopping it from revolutionizing industries—software included.