Good Enough to Replace You, Not Good Enough to Fix Itself

Here’s a new feature… There’s a new feature… Everywhere there’s a new feature. Can’t throw a stick around these parts without hitting a new ‘AI’ feature.

While I love the advancements, I do have to wonder if this is a ticking time bomb.

With so much speed in deployment and the increased vibe-coding adoption, how are all these ‘features’ being vetted before being released into the wild?

I’m no stranger to pumping out features with AI assistance, the key word being ‘assistance’. I do check code I generate before it hits the market, and some of it remains hand coded. That doesn’t guarantee 100% correctness, but I set the rules, goals, objectives, not AI. I have to understand what the code does, know how it works and feel confident in the choices made.

While there are many more like me, I do wonder how many more will become like me. As it becomes easier to lean on AI, it will lead to ‘brain rot’ or a lack of innovation. The chasm between technical developers and non-technical developers will continue to grow. If we become over-reliant on AI being at the helm, there’s likely to be a point where technology being implemented becomes so outlandish that only a select few could ever dissect what is going on.

When looking at the job market, we’re already seeing a problem accelerate towards us. It’s always been a bit of a beast, but there’s a division taking centre stage. All the technical junior roles are drying up, consumed by AI. So, no one can get experience. Meanwhile, employers are stating that there’s not enough technical talent.

How can anyone now get technical expertise if these foundation roles are been swallowed up? Experience is doing things, making mistakes, learning from the mistakes, trying something different and round we go. Break the cycle and it all falls apart.

This is one of those situations where I have to call a spade a spade - it’s cost saving. I’m sure others will go down the path of “it’s not about money, it’s about efficiency”. Don’t fret, I understand. AI is shiny and it’s great at doing repetitive pattern work. You could buy some super expensive frontier model credits (hello Fable), and it’ll still be cheaper than an employee.

Why would I spend more to get less? Why train someone when it’s so time consuming and costly, they’ll probably leave after a year anyhow taking that experience away.

Because of AI: “skip training the junior, we don’t need them anymore, load it onto mid-tier developers. Now they are equipped with AI, they will clearly become more efficient and have time for more work”.

Save like this today and it’ll cost you twice as much tomorrow. This noxious approach will almost certainly leave you holding the bag.

When AI goes wrong (and let’s face it, it happens) sometimes it’s simply not smart enough to figure out that its gone wrong. For new problems it has to rely on existing knowledge and in areas where little knowledge exists, you still need a way to provide that information.

There are so many tools being created right now to undo the very behaviour I’m referring to. A very simple example is writing. How many tools can you find that promise to improve ‘humanising’ what’s written. Groundbreaking idea… how about actually using a human?

Let’s be honest, for all the AI being deployed, are you seeing things becoming much better?

How about job satisfaction, has that improved since you deployed AI?

Do you have any staff left to ask that question to?

Have your costs come down?

From the perspective of ‘web pundit’, having visited a number of popular websites, they don’t feel faster (and I have a 1GB FTTP connection). They don’t even feel different, though in many cases I’m finding more bugs.

For all the new features people quack on about building, there’s a lack of any ground-breaking products as a result of AI. Not to say there aren’t any, but you would expect solutions to be whizzing into our lives at record pace. Maybe my glasses need cleaning, because I don’t see them. All this speed, all this output - and precious little of it is actually new.

Perhaps like me they’re all bogged down “reviewing” the code before it goes live…