We are not against A.I., but our stance is maybe more reserved that a lot of “a.i. bros”.
A.I. can be a useful tool, but it is over-promised and under-delivered in 98% of use cases.
A.I. is a tool, not the goal.
We think of A.I. of the tool that has to assist the human being of flesh and blood, instead of replacing him.
A.I. can be a useful tool during development for bug hunting and testing for example.
Problems with A.I. code generation
In our extensive tests, we’ve found some interesting downsides.
- Low effort
Would you bother to read an email you know was created with A.I.?
When the sender didn’t bother to write it, so why bother reading it?
So, why would you trust a company that didn’t bother to write the code themselves?
When A.I. generated code has proven to contains twice the amount of bugs than human written code.
Does that sound like a company that takes its customers seriously?
Would you trust the company that generates code with A.I.? - that A.I. doesn’t always understands the code it’s producing.
It generates code that simply doesn’t work as intended.
This is called hallucination.
But when you blindly click “accept”, then that is in your code now, causing bugs. - is it really faster?
You the programmer needs to be able to understand the code it produces, so you can hunt for bugs and unintended results.
Doing this is not only exhausting (because it eats a lot of brain space), it’s time consuming.
Because you did not think of the solution yourself, it’s harder to spot the edge cases that cause most bugs.
Often, it’s not even THAT much faster than writing the code yourself.
This is just one example.
We are not the only one experiencing A.I. coding takes longer.
An extensive study by Metr showed a slowdown of 19% by using A.I. when coding. - A.I. code generation makes you lazy.
With A.I. code generation it’s is really easy to fall into the trap of thinking: “it will probably alright”, and accept the code without thinking too much about it.
It’s extremely tempting to blindly accept generated code under time pressure, which is common in a lot of companies (time is money after all).
No programmer likes to read someone else’s code, especially when not even a human bothered to write the code.
Because you didn’t write the code, did not think of the solution, you have no mental investment or passion for the code or the project.
Studies have shown that a.i. generated open source projects are abandoned more quickly, than non a.i. generated projects. - Takes the fun out of programming
Programming is fun, because programming is the puzzle and the code is the solution.
But when you can copy/paste the solution, the puzzle isn’t much fun anymore.
Then why make the puzzle in the first place?
You can paint by number, but does that make you a painter?
With A.I. the programmer becomes an A.I. babysitter instead of a programmer. - A.I. generated content can’t be copyrighted.
Does that sound like a company with a vision?
Does that sound like a company that takes itself seriously? - It makes no sense
What is the point of using a software company to vibe code software, when you can vibe code your software yourself?
It makes no sense!
You paying a company to vibe code software has no additional value over doing it yourself!
You choose software like ours because of our human expertise and 25 years of experience.
Where A.I. is useful
- Extra pair of eyes
When A.I. is used as an extra pair of eyes for your code, or an extra brain to think.
This can be useful to find bugs, doing research or explore a concept. - Information resource
Because A.I. understanding code to a certain level, it can have a good grasp of your codebase is and what you are trying to achieve (context).
By asking A.I. targeted questions, it can give you answers tailored to your specific needs for a specific project.
It’s like having an extra programmer in the room that is there to help you. - Faster
A.I. is often faster than a question on a forum, or trying to find an answer on Google (that often doesn’t match your specific use case).
You also don’t have to go through the humiliation of 3 know-it-alls telling you that you are a stupid cow for wanting to do this without telling you why or giving you an alternative solution (you know … the typical Stack Overflow answer). - Prototyping
One of A.I.’s biggest weaknesses is also a strength.
When you want something quick that looks like it can do things, A.I. is perfect!
So making a prototype for an application, module or feature is a great application of A.I.
We don’t let A.I. write our code, nor do we force A.I. onto our users.
Bug hunting: yes!
Writing code: no!