AI Skepticism

I was lucky to graduate before AI

I graduated in ‘21, before AI became synonymous with engineering. It was around, there were majors in AI/ML. To me a naive young student, it was niche, and on the outskirts of what I thought was Computer Science. It was for people who paired their Computer Science degree with a minor in mathematics. It just never popped up on my radar, I was interested in frontend development. Things that you can interact with. After graduation the biggest hype was NFTs and Crypto. It was everywhere. Crypto was replacing central banks with DeFi. Fiat currency was dead. People were betting their entire life savings on these technologies. Not exactly notable, it seems like the general population has a propensity for gambling. What was notable was the fervor and consistency in marketing. Whatever industry you looked at, this technology was coming for it. It was going to replace everything. Whoever lived through this, you remember how absurd it all felt when there was a picture of a monkey selling for a million plus. Influencers doing rug-pulls every day of the week. It was the wild west. The thing is, it did not affect me. I didn’t gamble my life-savings, didn’t put everything into crypto, or whatever things were going on.

I remember what I was doing when ChatGPT came out.

Sitting at my favorite local coffee shop, working on something new for my personal site. I wanted to add a now playing widget to my website so anyone could see what I was listening to on Spotify. I was not familiar with the Authorization Code flow for API connections. I was reading through the documentation but not really getting anywhere so figured that I should check out this new thing everyone on Twitter is talking about. I signed up and dropped some questions in like “I’m trying to integrate Spotify’s web api into my personal website, can you help me with the Authorization code flow in sveltekit?“. I was blown away by the response I got, I immediately texted one of my best friends. “Have you tried out ChatGPT yet? It’s insane.” I think by today’s standards that response would be pretty lackluster. Just plain markdown, maybe some code snippets in code fences. It wasn’t very accurate and it was terrible at writing svelte code. Luckily svelte’s model was to be “just HTML/JS” with a little sugar here and there. I trudged through the prompts and wrapped my head around how to integrate with the API from svelte code. What I remember being so special was it materialized those conversations I could have in my head. I could ask whatever questions and get a good enough answer to either finish the rest by hand or be able to look up the right resources. I never felt like I got a major productivity boost. Just removed some friction in some places. It had replaced StackOverflow and traditional Google queries. Now by the time I ended up in those areas I had a much more pointed question to ask.

It used to be quiet

Once it had permeated through the initial users everyone always said, “It’s nice, it’s like a better Google search.” My parents started asking about it, eventually using it. It became a lot better; it was able to take multiple step reasoning, ask good follow-up questions and generate better code. It was everywhere. It felt nice, like it had found its niche to operate in. Then bigger numbers came out. Multi-billion number of parameters, terabytes of information that it was trained on. Some percent better than the last model and several percent better than all competitors. The signals became obvious, companies adopting AI features everywhere. It all became very noisy very quickly. Then a well-known AI researcher spent a weekend building something entirely led by an AI. Wrote about it and decided to dub this new technique as vibecoding. He didn’t invent it, but he sure gave it a spotlight. This is where the fracture happens in my mind. The work I had put into my degree and craft seemed like it was at risk. Everyone started saying the same thing. Software engineering is dead. This was my personal tipping point. It started to look very familiar. While I wasn’t able to see the internal side of the NFT/Crypto hype, I was currently in the middle of the next one.

I am very skeptical about these companies.

I have never seen such aggressive marketing in my life. Developers I used to look up to getting a check to shill the most recent model or tool. Super bowl ad spots, guerilla marketing. Multi-billionaires telling people that computer science is no longer a viable career path. These tools are replacing junior engineers. Laying off large amounts of their staff because one senior can orchestrate an entire team of agents. Companies fully shifting to AI first, bragging about how X% of their code is now generated by AI. It feels pervasive. It is one of those times I really feel like I am in a parallel world. I have used these tools. I have tried them all. I felt like every line of code I generated I lost so much understanding. I was the one driving but I was driving an F1 car when I wanted to be reading the road signs to make sure I didn’t miss my exit. I remember taking Claude Code for a spin after months of my co-worker swearing by it. It was going to replace us so we better get familiar with it. My first iteration of an internal tool. Simple on paper, maybe some architectural nuances but nothing crazy. It made something for sure but I had enough experience to see the bullshit a mile away. This was a classic instance of over-engineering and premature optimization. I thought it was my fault, I didn’t hold its hand enough and steer it correctly. Tried again, worse. Tried again, similar result. Got it to a working point. Went to run it and there were errors, attempted to debug it. Didn’t intimately know what it was doing so called on Claude again to fix it. It was junk. It was bad code that I didn’t know what it was doing. I deleted it all and wrote it from scratch. It worked exactly how I wanted it to. I knew what it was doing and what resources it was touching. That doesn’t mean I didn’t use AI. I used it a lot to plan certain sections and make sure I wasn’t doing anything stupid. It did help. It helped me to materialize thoughts and work through certain steps pragmatically.

I am not an AI skeptic. I think that it is a very useful and fundamental tool. It is here to stay, but I am anti whatever the fuck is going on in this industry right now. My principled take is to ignore 99% of the hype. I’ll keep working on the fundamentals and learning as a real software engineer. When everything settles down I will still be here without the rot associated with solely depending on these tools.