AI assistance has changed how software is written, and developers, including me, use it. The question for a client is what is still needed around the code.
Where AI tools are genuinely useful
- Prototypes and demonstrations to test an idea.
- Small personal or internal tools with low stakes.
- Speeding up routine work for someone who can read and judge the result.
Where the gap appears
- Understanding. Code you do not understand is hard to fix when it breaks, and it will.
- Security. Generated code can contain vulnerabilities, and handling login, payments or personal data requires review.
- Design. The shape of the data and the system is hard to change later and is not reliably chosen well by default.
- Testing and operation. Backups, monitoring, updates and handling real-world failures.
- Accountability. Someone who answers for it when something goes wrong.
A realistic middle path
Use AI tools to explore the idea. If it is worth continuing and real users or data are involved, bring in a developer to review, restructure and harden it, rather than starting again. Be open about how the prototype was made so the review can be honest.
The more a product matters to your business or your customers, the more the human part, judgement and responsibility, is worth paying for.