Does AI code well?
0 viewsI have been playing around with AI, especially Claude Code and ChatGPT's Codes for a while now and the anwer seems to be way more differentiated than a clear yes or no.
What AI can code
Standard tasks and repetive maintenance can be done to a decent level so far. That means yes, it can create a landing page that looks almost like every other landing page it has ever created.
Small step additions seem feasible, though the error rate is pretty high in terms of the thing coded actually working as expected. More often than not the AIs assume things or imagine things that do not exist in the codebase. For example adding a login form is simple, even if there are other fields, but don't expect AI to consider things like validation or the like out of the box.
What AI fails to do
I would summarise it as thinking. Yes, i know there are thinking modes, but the missing fragment is context evaluation the way we humans do it.
No human will create near identical landing pages, at least nobody worth their pay, they might be similar, but anyone who actually thinks finds different points to highlight and make visible depending on who or what the page is for. That is what designers and marketing departments do for a living and so far the AI version is underwhelming.
No human creates code that effectively copies functions from area to area within a single codebase. We have all learned that repetition is not a strength but a source of bugs. Anyone with experience will try to minimize code, not just add duplicates to add contradictions as potential future headaches. Yes, practically humans still add duplicates, but in my experience not at the same frequency as AI. This is in my estimate an over-focussing on solving a single issue at a time, leading to the AI not taking the whole picture into account.
No human with experience is going to completely ignore testing at any project that should last. AI is very conservative about writing test cases and running tests in my experience. If it was human, I'd joke it was afraid of being noticed messing around.
Closing Thoughts
While this is not optimal, the good news is, that we are looking at something like the industrial revolution. Low quality, mass products get automated while the high quality work stays with experts. So far I am not impressed by the output or speed of AI, but it is a useful tool for background work of repetitive nature. For example building a kanban board because I needed one with specific elements.
Regarding the Kanban board, I could have easily built it myself, likely significantly faster, but the benefit here was that I didn't and don't need a high quality output for now, just something that helps me organise my tasks in hobby projects. Exactly the kind of thing that you'd give to a junior engineer or an AI, if it works it helps, if it does not the harm is limited.
How has your experience with AI been so far?
Related Posts
- Is X a Growth Hack? — Software Engineering, 2026-04-07
- A question of databases — Software Engineering, 2026-04-08
- Blogging Experience - Or why i built a blog from scratch — Software Engineering, 2026-04-05
- Wolfgang AI - A Gamemaster in Trouble — Open-Source, 2026-04-14