Coding Mistakes to Avoid with AI

Can you build an app while watching your favorite show? Yes. Is it a good idea? Not always. Recently, a developer warned, “Did you let it build your own login system from scratch? It invented its own little password thingy. Looked real secure.” User's love for AI-powered coding tools is clear. They bring the twin promises of speed and ease to your fingers. Learning to leverage these tools correctly is crucial. “Use clerk or better auth. Roll your own and you leak every user on day one. This video walks through five core areas where the developer insists human oversight is essential: authentication, payments, design, database migrations, and copywriting.

Why “vibe coding” gets risky

"Vibe coding" is the act of letting AI do the heavy lifting in your software projects, often while you multitask. It is an approach that promises an easy and fun way to churn out software. However, by allowing AI to dominate the coding process, developers trade control for convenience. Risks abound when treating AI as a crutch.

Areas best avoided by AI (for now)

The creator highlights five key areas where AI assistance can be harmful. Like oil slicks and slick surfaces.

Authentication: The toughest nut to crack

Coding authentication systems from scratch is tricky. Mail and password handling can be complex. The process involves creating secure hash algorithms, storage mechanisms, and verification processes. AI tools can’t guarantee secure implementations, as they lack understanding of security nuances. They might invent their own “little password thingy.” You will likely put your users at risk. As the developer warns: “Roll your own and you leak every user on day one.”

Payments: The live grenade

This concern is particularly sensitive. Mismanagement of online payments could lead to financial loss and legal issues. The developer warns that refining payment processes requires caution. He elaborates on the risks involved in handling webhooks, refunds, and grant mechanisms. “refunds grants one sloppy line and you charge grandma four ground for a muffin”. With its precision, the video illustrates the consequences of letting AI handle money-related data.

Design: Your app's face to the world

Integrating AI in design could lead to a design that looks like yours. The developer believes that “AI Should Assist Design, Not Lead It.” Even if the final product looks good, there’s a risk it might not fit the brand. Allowing an AI tool to handle design without supervision may lead to bland, uninspired, or mismatched aesthetics. He points out, “You set the vibe, then let it help.”

Database migrations: An unsupervised disaster

Migrations involve shifting database schemas, updating data structures, and managing data integrity. It's complex. AI could easily make a mess here. Migrations can't be undone. Even errors such as data duplication or loss cannot be easily rectified. The creator puts it succinctly. “It typed a drop table. Super confident, though. That's how you delete production at 2 a.m. You donut. Data mistakes don't undo.”

Copywriting: Soul-sapping slop

Many AI tools can generate convincing human-like text, but they lack a deep understanding of context and nuance. They might not capture the essence of what you actually mean. Relying solely on AI for copywriting could lead to soulless, uninspired content. Your copywriting should be authentic and humanized. Anything else is a missed opportunity to be unique. But this developer notes, “It can write copy, but it doesn't know what you actually mean. So it just wrote, slot. Glorious soulless slop.”

Trust but verify. A practical approach

If you are looking to use AI coding tools, here’s some practical guidance:

• Become a Controlling Boss: Delegate tasks to AI but always review and verify its outputs.
• Take charge of critical areas: Ensure that the core areas of your application including authentication, payments, design, database migrations, and copywriting are thoroughly checked.
• Bridge the gap: Use AI to assist in non-critical areas to expedite the coding process, but maintain human oversight.
• Use AI tools to bootstrap your project, not run it: AI can help generate initial code, but it's crucial to refine and test it thoroughly.

The future: Living with AI coding assistants

The future is bright but also challenging. The increasing prevalence of AI coding tools will likely continue. Developers will benefit from them. However, they must remain vigilant. An ability to discern when AI can be trusted and when it can't is critical. In the ever changing landscape of technology, a little bit of wisdom, and a lot of guidance, can go a long way. Give it a try. Then step away from the keyboard.