How AI is upending the software development industry

It’s amusing to me to see the top SV companies falling back to the status symbols of yesteryear for juniors: prestige of your alma mater. Feels like an admission that they have no real way to vet candidates otherwise.

wishful thinking. Governments around the world are relying on algorithms to decide who can have welfare or who is cheating on taxes.

With the rise of AI it will only get worse.

The only thing it’s upending is sane engineering practices in my experience. I just reviewed a python flask app which was basically “vibe coded” by a couple people speed running a checklist of features. I open up the auth.py file to find it is integrated with an idp using oauth. Whether the caller has the correct scopes is never checked. I grab client credentials for a completely unrelated app. Hit /admin using my bearer token. Sure enough anybody with a valid set of client credentials could have had full admin access to this app for a year. But hey, at least when I asked how they tested the app they were quickly able to generate thousands of lines of pytest.

It’s cyclical. Always has been, likely to continue being so until an equilibrium is reached.

Leading up to the dotcom bust was a strong demand for technical talent. Talent that graduated in the midst of that bust got hosed, though were able to pick up work again as the economy realized that despite the bubble, technical expertise was still sorely needed.

Then came the 2000s, and the demand for technical talent skyrocketed on the back of Web 2.0 and the Digital Transformation of Business. Every college and grift was churning out the thinnest credentialed junior technologists to fill vacancies. Then the global recession happened, and suddenly every business started getting onboard the outsourcing wagon, thus crippling domestic technical roles and long-term IT careers. Except when ZIRP happened, suddenly everybody wanted software developers out of this notion that they could automate their way out of any conceivable problem with code.

Now LLMs are here, trained on the past and reducing the need for Juniors in those fields. The glut of C and D “bootcamp” students are superfluous as the business world re-orients around (comparatively) expensive debt and interest rates, and as technological revolutions effectively stagnate outside of a few, narrow fields (Quantum Computing, Cryptography, etc). AI coding tools built on LLMs are likely to decrease the roles of Juniors in orgs except for those high-quality candidates who work predominantly in newer languages or technologies LLMs can’t be effectively trained upon due to a substantially thinner corpus than, say, Javascript or C++.

Those of us who lived through prior cycles already see what’s coming next: refusing to hire Juniors ultimately cripples the pipeline of Senior talent, and educational institutions aren’t effectively training Seniors out the gate (since that requires both learned knowledge and lived experience). Seniors will likely appreciate in value in the short term (especially if LLMs prove my doomerism wrong and stick around/remain popular), followed by a business panic as they realize they footgunned themselves in the labor department by failing to hire, mentor, and train Juniors into Seniors. Combined with population growth declines, increasing retirement of older workers, and rising Nationalism clamping down on immigration or outsourcing, and you’ve got a perfect storm of sorts where companies will need talent but will have a lack of supply to pull from.

If I were a technologist whose background is bootcamps or vendor certifications, I’d be seriously worried about immediate and short-term prospects. Even if they hallucinate consistently and reward hack frequently, LLMs can supplant the kind of low-quality talent who normally take these roles in larger companies (at least for now). My advice is to start building more complex skill sets instead of relying on drills and rote memorization; contributing to FOSS isn’t enough when you’re competing against LLMs and thousands of randos for commit recognition, you’re going to have to make your own projects that solve problems or demonstrate your specific capabilities.

A and B-tier talent will have a rough few years as well, but they’re likely to soldier onward relatively unscathed unless I’m wrong and LLMs somehow lead to the Singularity. If you can’t get into that sort of quality, you shouldn’t pin your career hopes on a bootcamp or certification alone.

Companies want familiarity (repeatable processes) and “objectivity” (credentials). I always knew their spiel about “disrupting higher education” with bootcamps and free courses was just empty platitudes, and that once the party ended those who placed their faith in those schemes would be promptly abandoned in favor of the traditional Ivy League set.