A lot, actually. I developed an app using an AI assisted code but as the codebase grew, it became a messy ball of code and the tool started hallucinating.
Writing a prompt and getting code back is the easy part nowadays. The real work starts after that. And I learnt it the hard way.
- Did the AI coding tool understand the requirement correctly?
- Did it look at the existing code before making changes?
- Does the solution fit the current design?
- Did it break anything else?
- Were the right tests added? 6 Does the developer actually understand what changed?
This is the part that often gets skipped in AI assisted coding.
The prompt gives you a starting point. The pull request and merge is the final outcome. But between the two, you still need the same engineering discipline: understand the problem, explore the codebase, think through the design, implement carefully, test it, review it and check the impact.
AI can make coding much faster. But faster code is not always better code.
The real question is not just “Can AI write this?”
It is:
“Can I understand, validate and confidently own what AI has written?”
That space between the prompt and the pull request is where good engineering still matters.
Top comments (0)