There are low difficulty tasks in programming.

There are low difficulty tasks in programming.

https://gyazo.com/7461b44afcb20f544f39cd08cd0660f7

Programming is like this

Small black circle: Do light work that you know how to do (git clone, npm install, etc.)

Green medium circle: with the expressive activity of "implementation".

Until now, programmers have been doing this process all by themselves, so there was not so much frequency of solving the riddles.

Especially when it comes to team development, the frequency has decreased even more because there are more black circle tasks such as writing specifications, cleaning up commit logs, writing documents, requesting reviews, reviewing, etc. to share information with the team.

AI agents behave as if they can execute some small rounds.

The size of the green circle to be implemented depends on human instructions.

https://gyazo.com/3e4ac60ce0d9a0d3f4a453b9a2c425ed

A: The person with the right sized instructions will say, "Wow, great, you did all the work in one step!" And it goes.

B: Those who gave too big instructions would say, "You can't implement it at all.

C: For those in between, the AI says, "I've got it implemented!" so I check it and it's buggy.

So we have to do all the debugging, which is very difficult for humans.

I felt the picture wasn't appropriate after writing all this.

The AI doesn't notice the bug, thinks it's complete, and proceeds to the end.

Adding tests increases the power to detect bugs.

People in the age of AI assistants start out with only red-circled riddles to solve, and then learn to go down the abstraction layer out of necessity to solve the riddle.

---