Ask HN: SWEs how do you future-proof your career in light of LLMs?

Without doxxing myself or getting too deep into what I actually do there are some key points to think about.

For very simple to mid complex tasks, I do think LLMs will be very useful to programmers to build more efficiently. Maybe even the average joe is building scheduling apps and games that are ok to play.

For business apps I just do not see how an LLM could do the job. In theory you could have it build out all the little, tiny parts and put them together into a Rube Goldberg machine (yes that is what I do now lol) but when it breaks not sure the LLM would have a prompt big enough to feed the entire system into to fix itself.

Think of this theoretical app.

This app takes data from some upstream processes. This data is not just from one process but several that are all very similar but never the same. Even when from the same process new data can be added to the input, sometimes without even consulting the app. Now this app needs to take this data and turn it into something useful but when it doesn't it needs to somehow log this info and get it back to the upstream app to fix (or maybe it's a feature request for this app). The users want to be able to see the data in many ways and just when you get it right, they want to see it another way. They need to see that new data that no one even knew was coming. To fill in the data that comes in this app needs to talk to 20 different APIs. Those APIs are constantly changing. The data this app takes in needs to send it to 20 different APIs through a shared model, but that model also takes into account unknown data. The app also sends the data from the upstream process to native apps running on the users' local machines. When any of this fails the logs are spread out over on-prem and hosting locations. Sometimes in a DB and sometimes in log files. Now this app needs to run on-prem but also on Azure and also on GCP. It also uses a combination of Azure AD and some other auth mechanisms. During the build the deploy process needs to get the DB, Client Secretes and roles out of the vault somewhere. Someone needs to setup all these roles, secrets, ids. Someone needs to setup all the deploy scripts. Now with every deploy / build there are automated tools to scan for code quality and security vulnerabilities. If they are found someone needs to stop what they are doing and fix the issue. Someone needs to test all this every time a change is made and then go to some meetings to get approval. Someone needs to send out any down time notices and make sure the times all work. Also, the 20 apps this one talks to are the same as this app.

I could keep going but I think you get the point. Coding is only ¼ of everything involved in software. Sure, coding will get faster, but the jobs associated with it are not going away. Requirements are crap if you even get requirements.

The way to future proof your job is get in somewhere where the process is f-ed up but in a job security type f-ed up. If your job is writing HTML and CSS by hand and that is all you do, then you may need to start looking. If your job has any kind of process around it, I would not worry for another 10 – 20 years, then we will need to reassess, but again it is just not something to worry about in my opinion.

I also know some people are building simple apps and making a living and the best thing to do is embrace the suck and milk it while you can. The good thing is if LLMs really get good like good you will be building super complex apps in no time. The world is going to have a lot more APPs that if for sure, will anyone use them is the question.

If developers become more productive (both juniors and seniors), there will be more demand for all of them.

I recently used an LLM to translate an algorithm from Go to Python at my work. The translation was quite accurate, and it made me think tasks involving obvious one-to-one correspondence like code translation might be easier for LLMs compared to other tasks. I can see the potential for offloading such tasks to LLMs. But the main challenge I faced was trusting the output. I ended up writing my own version and compared them to verify the correctness of the translation.

> LLMs are barely at the level of a diligent but not very good junior dev at the moment, and have been for the last at least 6 months.

> Your take is rather romantic.

I’m not sure you’re aware what you’ve written here. The contrast physically hurts.

My advice: Keep honing your problem solving skills, by doing math challenges, chess puzzles, learn new languages(not programming ones, though that might help too), read books; anything that’d help you get newer perspectives, challenges your mind is good enough to withstand the race against AI.