Has AI made me dumber?
An anxious girly’s diary entry on tech & self-esteem
Friday afternoon, as I was logging off after a long day of telling Claude what to do, I ✨seemingly✨ out-of-nowhere started to reflect on how different the challenges I face in my work are now compared to three and a half years ago, in the pre-ChatGPT era.
But before I expand on my reflections, let me give you a bit of background about me and about what to expect from this post. I have been working as a Software Engineer for almost four years, and I have a BSc in Computer Science. In this post, I am not going to defend a pro/anti AI sentiment, nor will I explore ethical, scientifical, or environmental considerations surrounding AI. Those conversations are important and I encourage you to dive into them - but here I will simply ramble on how AI has shaped my work, and how it shaped how I feel about myself.
So how exactly has my work as a software engineer changed since the boom of LLMs?
First of all, I am writing much less code. Mind you this does not mean I am shipping less - on the contrary. LLMs have made it incredibly easier to generate new code. At first, in 2023, they were mostly only good at creating template code - for example unit tests or test data. So the split between the human and the machine was truly like this: let the LLMs do the repetitive boring stuff, while developers continue to do the reasoning and innovating.
But I don’t need to tell you that since then they have gotten much better. Now it’s enough to list well the requirements of a task, and have a general idea of what your code needs to do, and voila! An agent can do the work for you end-to-end. This means that as a developer you need to spend less time familiarising yourself with a new codebase. As long as you know what needs to be achieved, you can leave it to the agent to understand where it needs to plug code changes. Getting started with a new codebase used to be quite time-consuming, but AI tools have made it so that you can be in-and-out of an unknown codebase within the day, and with your PR merged. This has helped me get things done significantly faster. But it has also deprived me of the need to understand things in depth. In a way, instead of me defining the context window of an agent, the agent is defining my context window: this is what you need to know, human - leave the rest up to me.
However, one shouldn’t fall into the trap of believing that just because a process is longer or more tedious, it’s necessarily better or more worth-while. Why walk from London to Paris when you can take a train? When it comes to getting your work done in a company, not needing to burden yourself with convoluted business details can be quite desirable. Spending 5 days deciphering a long-gone coworker’s work from 10 frameworks ago just to change one line doesn’t exactly sound rewarding.
But there’s another catch - a good developer must strike the perfect balance between what needs and what doesn’t need her attention. You put too much distance between your context and the details the AI can deal with, and you are no longer able to tell when the agent is making sub-optimal decisions. (Have another LLM review in adversarial mode, some would say …). You are also going to have a much harder time returning to a task after a while to build something else on top of it. What I’m getting at is quite intuitive - doing a task yourself will always make more things stick with you than reading someone else’s “here’s how I did this” report. But like I said before - not all information at your work is worth collecting.
LLMs have also made it so roles are becoming increasingly tech-agnostic. The conversations are no longer centred on stack - should we use Python or Go, and what developers have experience in what programming language? Instead, when a task needs to be done, the developer is expected to just jump in whatever programming language is required for it. An LLM can handle syntax, conventions - everything, really.
But why be sad when you’re doing more?
I started coding in high school, and one of the things that convinced me to study computer science was a website I built during a summer camp. The website was supposed to allow you to build your own virtual flower bouquet, with recommendations, and order it - had this ever become a business idea. It had a rudimentary username and password login, stored unencrypted in a database on my laptop, and a girly flowery UI that took me days to put together. I remember feeling so empowered when things started working. I felt that I don’t just know things, I do things.
Now this kind of project would put GitHub Spark to work for probably 20 minutes. At the end of those 20 minutes, everything would be fully built, and you could choose between 10 different colour themes. So if I ever need to build a virtual flower shop website - or, let’s say, a script that builds a pie-chart for my wins and losses during my chess streams, there’s no need for me to spend days building it. (But if I wanted to, I still could, right?) I just need to use my computer science knowledge to have a vague idea of how things need to be done, prompt a model in a clear way, and poof! Magic!
This has given me a lot of energy on good days. I’d feel what these tech companies want you to feel: that I can ✨ DO THINGS ✨. I’ve never done baking? Let me ask ChatGPT for a recipe, tips, and double-check whatever insecurities I have about the texture of the dough. I want to experiment with the Lichess API? Let me work with Claude for a couple of hours on a Sunday morning. I want to improve my general knowledge? Let me task GitHub Spark with making a small app that sends daily facts in my inbox.
But back to last Friday … what happens when you start fearing that while you’re “doing”, you’re not actually… doing?! What happens when your brain starts wanting to just skip that middle step? When it doesn’t want to read through the ten-paragraphs response of ChatGPT and just wants to click “Start implementing”? What happens when what needs to be done requires you to be not much more than a distant manager of the task, so your brain just wants to skip to the completion stage already? I guess this is when I started worrying that I have lost my sharpness. It doesn’t help that the leading diagnosis on the Internet is that too much short-format content melts your attention span. And not needing to fully engage and swapping between tasks while agents are working - that doesn’t sound like something that would preserve your attention span either.
And this is why on Friday, despite wrapping up my to-dos for the day, I didn’t feel good about myself. I felt that despite getting more done, my brain is actually working less. ( Well, having a culturally induced strong link between self-worth and productivity is also at fault here, but let’s talk about that some other time.)
So, have I gotten dumber?
Um, probably not. I did just write this piece of text that you kindly read without touching my phone for like 2 hours. (Wow! Amazing! Increible!) And also there isn’t really an empirical way to measure my dumbess or its absence. But the fear that my cognitive abilities have atrophied needs to be addressed, while staying pragamatic about working a job.
I told ChatGPT about what I was experiencing 😂. It assured me that I’m not the only one having this anxiety. It gave me the following advice: learn a new language, work on a side-project without AI, and… just chill, girl. As I tried to do the Claude-prescribed yoga exercises to de-stress, I managed to glance deep into my soul and came up with the perfect solution. And the best part of it? It was right under my nose the entire time.
How to feel less dumb
Imagine an activity where using a phone is completely prohibited. Where even the mere sight of an electronical device is severely frowned upon. An activity where all you have, for hours, is your own mind, working, scheming, bending itself through time. An activity that can give you the satisfaction of feeling like a genius, of doing something hard, only to humble you right aferwards and compel you to want to do better. Guessed it already?
The cure to the epidemic of short attention span (real phenomena or anxiety, works for both)
It’s classical chess, babe! 💁🏻♀️ No better brain workout than staring at small pieces of wood for hours, overthinking their possible movements again and again. If your brain can go through that, you’ve definitely not gotten dumber! So see you at the chess board … 😌
