Where I Use AI Coding Assistants in Real Projects, and Where I Don't

Do these tools really change your work?

In 2026, "writing code with AI" is no longer an experiment; it's part of the daily job. But there's a big gap between the impression marketing material gives and real project experience. I used these tools heavily while writing PlusTurk's collections module and Ayda's kitchen display sync, and I saw both where they genuinely save time and where they mislead.

This isn't a theoretical comparison. It's what I learned while building a real SaaS product and a desktop app.

Where they really speed things up

Repetitive CRUD and form code. For "I know how to do it, it just takes time to write" jobs, like admin panel forms, API routes that follow Prisma schema changes and form validation, AI assistants genuinely save hours. I sped up most of the form components in this site's admin panel this way.

Writing tests. Thinking through the edge cases of an existing function and writing test scenarios is an area where AI is more patient than a human developer. You set up the logic, and you finish the tests together.

First contact with an unfamiliar library. Reading a new payment provider's documentation (Nestpay, for example) and producing a first integration draft is much faster with AI. But it's a draft, not code that goes to production.

Where I don't trust them

Architecture decisions. I don't leave decisions such as how to build data isolation in a multi-tenant SaaS, or how to normalise which table, to AI. When these decisions turn out wrong, you notice months later and fixing them is expensive.

Security-critical code. I review code that handles payment flows, authentication and permission checks line by line myself. The chance of AI producing code that "looks like it works but has a security hole" is real, and you can't afford that in a system like PlusTurk where real money moves.

Knowing the answer to "why did I do it this way?" If you're going to sit down with a client and explain why a feature works the way it does, the reasoning behind that decision has to be yours.

A practical rule

In my own work I follow this rule: AI gives me a draft, I understand it and own it, then I deliver it. I don't hand a client code I don't understand, because I'm the one who'll support it, not the AI.

// Example draft from the AI: a safe middleware, but one I still need to review
export async function withAuth(req: Request) {
  const session = await auth();
  if (!session?.user) {
    return new Response("Unauthorized", { status: 401 });
  }
  // I added the role check; it wasn't in the draft
  if (session.user.role !== "ADMIN") {
    return new Response("Forbidden", { status: 403 });
  }
  return null;
}

Conclusion

AI coding assistants don't replace a developer; they change the tools in an experienced developer's hands. The difference is knowing which job to hand over to whom. I use these tools because they speed up my work, but the decision, the responsibility and the promise I make to a client are still mine to carry.