Why AI (probably) won't take your job

There's absolutely no doubt that AI has absolutely ingrained itself into our lives and shaken up the software industry. If you haven't been following along with the advances in AI technology you're doing yourself a disservice, but is AI going to replace you?

Copilot

When I first tried copilot, it was an awakening to be taking AI seriously. The joy that copilot brings is not something that can be replicated by any other tool.

Since the initial release of copilot, we've seen AI make supposed massive improvements. Every new model is better than the last etc. etc.

But has this really been the case? Sure, copilot is better than no copilot, but is the copilot of today meaingfully better? I don't think so.

Copilot shines at being a glorified autocomplete. It certainly makes writing code easier, but it's not a magic wand that can solve any problem.

I especially enjoy using copilot for writing code in unfamiliar languages. What used to be a search through Ruby documentation is now just given to me in an instant without leaving my editor.

But all to often, the suggestions are at best not what you want and at worst full of issues. I would hate to be a junior engineer at the moment because you wouldn't know what to do with the suggestions. Fundamental knowldge is hard to learn if you're reliant on copilot.

For all level of developer, I think having a break from copilot is a good idea to ensure you're not always relying on it to solve your problems.

Prompt engineering

I have dabbled in trying to use prompt engineering to solve larger problems. I attempted to use v0 for my Thai time converter tool . While I got given a basic UI, the underlying methods to convert the time were not accurate at all, so I still spend a significant amount of time trying to figure out how to get the correct output.

This really sums it up. AI may get you 90% of the way there, and you either don't realise it's wrong and you're shipping a product that doesn't work, or you're spending time on fixing it. There's definitely time to be saved, but using AI for 100% of problems doesn't seem like it will save time in the long run.

Shortfalls

The thing I would love to see is an AI that can deeply understand a large codebase. All of the solutions I've seen so far don't seem to hit the mark.

The AI agents advertised to perform code reviews only seem to be catching simple mistakes and I've seen some advertised that show the AI picking up simple linting errors, easily accomplished with existing tools.

It has been nearly a year since Devin was announced, and similar tools that were touted as a replacement for humans, and we're yet to see any real world impact. Even Devin is now marketed as an &qout;AI teammate" as opposed to an actual replacement (a much better strategy!).

I've been pretty impressed with Supermaven as it seems to have a large context window indicating it has a decent understanding of the codebase, so it seems possible that we will see improvements in this space.

Can anyone be a developer now?

For all of the above reasons, I think the answer is no. But I do think the landscape is changing.

If you split the software world into engineering and product, I think we're going to see the roles converge. AI is good enough for a non-engineer to get started on a project. But they will need to have a solid understanding of how to control an AI to get the outputs they want. This kind of understanding is an engineering trait. So while the barrier to entry is now lower to become an engineer, you'll still need to have a decent amount of expertise and be able to think like an engineer to see good results.

For engineers, I think the best way to ensure you're valuable is to deeply understand the problems you're solving. This is true with or without AI. Don't just be someone who writes code, be someone who can solve problems. Solve problems for your users. In other words, learn to be a product manager. Learn to be a designer.

There will also be a day in the near future where AI generated startups are in an absolute nightmare state and engineers will have to clean up the mess.

One should not bury their head in the sand and hope for AI to go away, the cat is out of the bag. Embrace the change in the industry and use AI as the great tool that it is, and make sure you're providing value in ways above and beyond just writing code.

I will be interested to see how this ages...