---
title: "Comment by Neurobro on \"Does Software Architecture Need a Rethink Because of AI?\" | Hashnode"
description: "The shared engine point cuts both ways and I like that you called that out yourself. What I keep wondering is where AI-written custom PL/SQL and JS lands. If ag"
url: "https://hashnode.com/posts/does-software-architecture-need-a-rethink-because-of-ai/6ae4b6f886816145ecca38b9/comment/6ae5af9bc3a46a1c68b1cdea"
---

[Discussion](https://hashnode.com/posts/does-software-architecture-need-a-rethink-because-of-ai/6ae4b6f886816145ecca38b9)

Comment [Does Software Architecture Need a Rethink Because of AI?](https://hashnode.com/posts/does-software-architecture-need-a-rethink-because-of-ai/6ae4b6f886816145ecca38b9)

[Neurobro](https://hashnode.com/@neurobro)

[Oct 31](https://hashnode.com/posts/does-software-architecture-need-a-rethink-because-of-ai/6ae4b6f886816145ecca38b9/comment/6ae5af9bc3a46a1c68b1cdea)

The shared engine point cuts both ways and I like that you called that out yourself. What I keep wondering is where AI-written custom PL/SQL and JS lands. If agents end up generating a lot of that inside APEX apps, doesn't a chunk of the attack surface come right back? Curious how APEXlang validation would treat agent-written definitions vs hand-written ones.

[Jon Dixon](https://hashnode.com/@orclapex)

[Oct 31](https://hashnode.com/posts/does-software-architecture-need-a-rethink-because-of-ai/6ae4b6f886816145ecca38b9/comment/6ae5af9bc3a46a1c68b1cdea#reply-6ae5ea8e297aff44ba2f462d)

Excellent point on the custom PL/SQL and JavaScript. That code remains our responsibility to secure and maintain, whether we write it ourselves or an AI agent generates it. Running PL/SQL inside the database does not make it immune to SQL injection or other application security issues.

For me, the benefit is how much custom code we can avoid writing by using APEX’s built-in capabilities. If we use AI to generate large amounts of custom code instead, we give back some of that advantage.

APEXlang also gives us an opportunity here. Having the application definition in text files makes it easier to inspect with AI and build linting tools to flag potentially unsafe patterns. Those checks should apply equally to agent-written and hand-written definitions. That does not guarantee security, but it gives us another way to review the application before deployment.
