The new price of entry to software
Summary
In a future where no one writes code anymore, proprietary AI subscriptions may raise the barrier to entry for the software industry.
I grew up in a middle-class family. My mom was a military nurse, and my dad was a military engineer turned stay-at-home dad. We weren’t poor, but as a kid in newly liberalised India, I didn’t have the same access to technology as Western kids my age. My first exposure to computers was in a school lab, where we learned BASIC and FORTRAN.
We didn’t have a computer at home until I turned 20. When my parents agreed to fund my first computer, I had to assemble it by hand. It was an adventure to go shopping for parts, put the computer together, install the operating system, and then wait for the magic that happened when the system booted up. My first computer had two partitions. On one I ran Windows 98. On the other, I ran Debian Linux. It unlocked the power of the UNIX terminal and helped me learn to code. Between labs and my MacGyvered computer, I had everything I needed to pick up skills to land an IT job.
Truth is, many of my friends didn’t have a computer at home until a few years after they began working. Mere access to a lab was enough to learn computing. The barrier to entry to software careers was low. Open-source software democratised that access further.
As luminaries like DHH announce the end of coding as we know it, I wonder how a poor kid in India learns the skills to join the software industry. You still need access to a lab or a computer. But beyond all that, you need access to AI models. Frontier models and their proprietary harnesses are costly to access. In many Indian towns, a $200-per-month Claude Max license costs more than the rent for a two-room apartment. Even the $20 licenses are often out of reach for some families. While open-weight models are matching the frontier models’ capabilities, hosting an open-weight model and pairing it with a suitable harness is a daunting barrier. Suddenly, it’s not that easy to get the exposure you need to be a software engineer.
Now, there are many ways to view this situation. For several years, the barrier to entry for software engineering was low. You didn’t need to go to college. If you were curious, a hand-me-down computer in your bedroom could set you up for a career in tech. You couldn’t do that with, say, civil engineering. You’d need exposure to high-end equipment to gain meaningful experience. Perhaps that’s how it’ll be for software engineering, going forward. To learn how to be a software engineer, you may have to attend a college with a well-appointed lab and work for a company that provides you with a handsome token budget. If no one is hand-chiselling code anymore, perhaps that’s a good thing. There’s a difference between knowing how to code and knowing how to engineer software. Perhaps the future belongs more to those trained engineers than the self-taught coders.
Or perhaps the future also belongs to people who will enter the software industry, not because of their technical skills, but because of their right-brained competencies. Barriers to learning design, storytelling, and empathy remain low. If the effort to build software is collapsing, the industry could use more right-brained competencies to decide what to build and cope with the entropy that comes with AI-powered software development.
Or perhaps I’m speculating too soon. Maybe inference on the edge will get so good that we won’t have anything to worry about. Hosting models on a personal computer is resource-intensive today, but they needn’t be in the future. Perhaps that poor kid, with a hand-me-down laptop or a Raspberry Pi-powered desktop, will be able to build software using an on-device open-weight model and an open-source harness.
“Prediction is very difficult, especially about the future.” — Niels Bohr
Whatever the future may hold, the industry faces a junior engineer crisis today. Teams are getting smaller and more senior. The junior year wall is real. Junior engineers don’t have as much time on task to learn effective engineering patterns. Junior-level entry has gotten harder, and salaries have stagnated. A career in software isn’t as attractive as it used to be. Meanwhile, it is also true that without a junior layer today, the industry won’t have a middle and senior layer in the future.
Many competing hypotheses exist about the future of software development. Some believe that we’ll see a seniorisation trend in software teams, and the teams will be smaller and closer in physical proximity to the businesses they serve. As a result, the industry will need fewer software developers. Others believe that we’ll see a Jevons paradox-style explosion in the demand for developers. If building software becomes cheaper, it could increase the demand for developers. Whichever speculation comes true, the industry can’t ignore how it sources, attracts, trains, and onboards junior engineers. I hope we address these questions sooner rather than later.