It is a very difficult time for software engineers at the moment. Whether you like AI or not, it is having an impact on the software engineering job market.
Some companies are using AI as an excuse to layoff more staff, either because they over hired or because they are cutting expenses in order to spend more on AI.
There are even lots of reports of companies demanding their staff use AI for every task. I do agree to some extent that AI can make you more productive in the short term. I have certainly been able to complete more tickets in the last few months than I would have been able to do otherwise.
The use of AI does however come at quite a steep long term cost. Not only are developers forgetting how to write code as they rely more on AI, they are also no longer are able to keep a mental map of the codebase which is critical for debugging issues. Over the next few years we are going to see software get buggier, companies having more frequent and longer downtimes and internally codebases are going to get messier and harder to maintain.
I am hoping that companies who are currently forcing AI use will have a U-turn when they notice the long term impact and be more selective in how AI is used going forward.
If that doesn't happen however, what skills are still going to be relevant over the next few years that are unlikely to be replaced by AI?
This is what I still think is relevant now and will be going forward:
Testing #
If companies end up using AI to write all the code, then you still need to test it, one way or another. Of course, you could just get AI to write the tests which for the code itself might be fine, but you need to define what to test yourself.
I can see the AI apocalypse going one of two ways. AI causes more bugs in the software, and it is starting to affect the companies bottom line and reputation.
- The company back peddles on using AI everywhere, and we go back to coding manually at least partially.
- The company doubles down and wants you to use AI to write more tests.
Either way I think we are going to be expected to write more tests to ensure the AI code is doing what we think it should be doing.
If AI is writing the majority of code then I doubt unit tests are going to be that helpful. The majority of the effort is going to be writing integration tests for APIs and automated UI tests for the frontend.
System Design #
AI is great at adding features to an already well-organised codebase. However, if you use AI from the start there is no code structure for it to follow.
Knowing what good software looks like, how to structure the application so that it will be easier to maintain going forward is going to be very important.
This extends to the system as a whole as well. AI might be able to suggest the system design you need but unless you make it a requirement while you are coding, it won't be considered.
Ultimately, knowing which database to you use, when to introduce queues for certain tasks and understanding how to scale different components is your responsibility.
Debugging #
When trying to fix a bug with an unfamiliar codebase I find myself using debugging a lot. If you didn't write the code then you don't have the same mental map of the codebase which allows you to easily work out the issue. The only option is to go line by line and look at what data is entering each of the functions, to see why the code took a certain path.
This also means looking at the logs and not just asking Claude whenever an issue comes up.
Problem Solving #
Software engineering has always been about problem solving not code. You need to get a full picture of the problem and understand how code could help and more importantly when it isn't needed. Everything can't be solved with code, but AI is definitely going to try.
Every line of code is a liability and the best line of code is the one you didn't have to write. AI will have no problem writing code for a feature, but it won't ever stop to think whether the feature should have existed in the first place.
❤️ Picks of the Week #
🤖 AI #
Changes in the system prompt between Claude Opus 4.6 and 4.7 - It is interesting to see how much the system prompt changes. I am not sure how we are supposed to rely on these systems if they fundamentally change so much.
Atlassian enables default data collection to train AI - Are we all just going to have to self-host our own tools from now on. For my personal projects I just use Obsidian with the Kanban plugin.
OpenAI Privacy Filter - I am glad this is an open weight model as it you shouldn't be sending any PII data to a third party.
Talkie: a 13B vintage language model from 1930 - This is fun. I am not sure how useful it is but fun anyway.
Who owns the code Claude Code wrote? - This is an interesting one. If code generated by AI is deemed uncopyrightable (is that a word?) then surely companies will go back on using it everywhere. Given it is trained on stolen code I can see how it wouldn't be.
AI is making me dumb - I never use AI to write these newsletters or any of my blog posts. For code however, with work almost mandating the use of it, it can be so tempting to defer to it for everything.
I believe there are entire companies right now under AI psychosis - AI is going to create a whole generation of dumb engineers that don't know how to think or code. Maybe that is what they want so companies have to pay us less.
SANA-WM, a 2.6B open-source world model for 1-minute 720p video - This looks cool, but I dread to think of the compute needed to generate these. I have tried video generation locally on my M2 Pro Mac Mini, and we are talking 30 minutes at CPU melting full speed to generate 5 seconds of questionable slop.
How fast is N tokens per second really? - For those running AI locally they often talk about tokens per second, but it is hard to understand how fast that is without actually seeing it. This helps a lot.
The last six months in LLMs in five minutes - There is so much happening in the LLM space right now. This is a good summary.
Changes to GitHub Copilot individual plans - The era of subsidised frontier LLMs is slowly coming to an end. Let's hope the local models can compete before it gets too expensive.
GitHub Copilot code review will start consuming GitHub Actions minutes - I wonder how many companies are using Copilot review as a replacement for Human review.
🛠️ Tools #
Flipper One – we need your help - The Flipper Zero looked so cool when it came out. I didn't get one as I don't know what I would use it for. Equally, the One looks very cyberpunk, and I want one, but I don't know why.
Framework Laptop 13 Pro - I am likely good for computer hardware for at least another 5 years, but I can see my next computer being from Framework. I really don't like the way that Apple is heading.
💻 Programming #
Laws of Software Engineering - I think software engineers that actually understand these “Laws” and can apply them are going to become so valuable in a see of vibe-coders.
Your hex editor should color-code bytes - This is very cool. The first time I played around with hex editors was with a GameShark Cheat Cartridge for the N64.
The seven programming ur-languages (2022) - I never realised all the programming languages I know fall under the “ALGOL” family of languages.
🔒 Security #
GTFOBins - I have started running my AI coding agents inside a VM using Lima. It might protect against these types of exploits but who knows!
Click - It is amazing to see how much information can actually be gathered from the browser. Even your mouse movements can be recorded, it is quite creepy.
🎨 Design & UX #
How to make your text look futuristic - This is hilarious, why do we always pick the same styles for "future text".
Moving away from Tailwind, and learning to structure my CSS - I moved my website over to Tailwind last year after using it a lot at work. I do feel like my HTML is now very bloated though.
Plain text has been around for decades and it’s here to stay - The plain text UI tools mentioned here are so cool. More people should start with ASCII wireframes.
🏢 Tech Industry #
Drunk post: Things I've learned as a senior engineer - This is so true, especially point number 1.
The best way I’ve advanced my career is by changing companies.
Software engineering may no longer be a lifetime career - I definitely think my days of software engineer as a career are numbered. I will still make software just not for someone else. I don't mind using AI selectively on my own projects, but I don't want to be forced to use it.
Incident Report: Railway Blocked by Google Cloud (Resolved) - I spoke about Railway last issue as a quick way to get your own ForgeJo instance started. It is still the quickest way but I wouldn't rely on it for your company git repository.
Google just spat in my face - It is really tough to be a web developer right now. I once thought I could always go freelance and build websites for companies, but I think that path is slowly dying. At least Google is doing it's best to make it that way.
🎮 Gaming #
DOS Zone - This is so cool. They have SimTower on here as well. I played that so much as a kid. It had to be installed using 4 floppies if I remember correctly.
📚 Books #
Pinocchio is weirder than you remembered - I never knew Pinocchio was quite so creepy. The donkey bit in the Disney movie was weird, but the book is on another level.
Technical Reading List - There is some good engineering reads on this page. OK there are only 3 books mentioned, but I felt it fitted in this category.
🧠 Miscellaneous #
Period tracking app has been selling data to Meta - Just the sort of information you don't want going to a big tech company who is known for using your data to sell ads. I am probably not the best person to make one of these apps but I feel like make a local alternative just so people have options.
Flipdiscs - These looks so cool. I am not sure I have the space for an art installation in my office but if I did!
💡 Today I Learned (TIL) #
I was having issues after killing a command which left my terminal writing ^[[A and ^[[B whenever I used the arrow keys. It turns out you can type stty sane to fix this!
💬 Quote of the Week #
How do you cultivate discipline? By building habits – starting as small as you can manage, even microscopic, and gathering momentum, reinvesting it in progressively bigger changes to your routine, and building a positive feedback loop.
From the article "Screw motivation, what you need is discipline".
Top comments (0)