Core Competencies for Expert AI Users · The Attitude We Bring to Speaking on Social Media · Mutual Distrust of AI-Made Work
For ordinary people developing software with AI, security problems are, in principle, solvable. You can simply ask the AI to fix them.
The real problem is that AI may proceed without recognizing that what it is building is intended to become an actual production product. At the same time, the person prompting the AI may not even realize that there are things they do not know—so they have no reason to provide the missing context in the first place.
That creates blind spots that neither side ever touches.
And this can happen even with source code the AI can directly access. What, then, happens with infrastructure that is much harder for it to see?
In principle, the problem can be addressed by giving the AI careful and comprehensive context about the infrastructure. But that is precisely where the difficulty lies: we generally cannot expect an inexperienced developer to know what context needs to be provided, much less to provide it thoroughly.
Nor is this limited to security. The same problem can arise with databases, networking, and other parts of a system. Take database migrations, for example. Even when the database-related code is fully accessible to the AI, I suspect it will become increasingly common for AI-assisted app development to overlook a migration and end up wiping out user data.
If saying something correct still provokes resistance, before blaming the other person, it may be worth asking whether there was also a desire to defeat them—to place yourself above them.
After all, the way we speak has a tendency to reveal the desires that run deepest underneath it.
Just as developers do not fully trust code written by AI, users do not fully trust apps and services built with AI.
More specifically, when developers themselves become users, they are often among the first to wonder whether some service casually built by an individual with AI will actually be maintained, and whether its features have ever been properly tested.
Ordinary users may not be so skeptical yet. But once they have accumulated enough experience with AI-built apps—and learned from that experience—they will begin asking the same questions.
One of the things to be most wary of on social media is the inability to resist the urge to speak even when there is nothing in particular worth saying.
