How Not to Use AI? Facts Through Experience

Real experiences on using AI: coding, research and risks. If you are not in control, AI will manage you.

How Not to Use AI? Facts Through Experience

AI use is different for everyone. This article does not tell a “truth”; It contains only my observations and experiences.

Today, AI is generally used for 3 purposes:

  1. Writing code
  2. Research
  3. Entertainment

1. Writing Code

Starting a project has become very easy with AI.
Scaffold, structure establishment, first deployment → provides significant speed.

But the real problem starts after revision.**

Problem

There are two critical scenarios:

a) Starting the project from scratch with AI

First version:

  • Works nice
  • The document is correct
  • It is deployed

But then:

  • New needs come
  • Revisions begin
  • Context disappears

Result:

  • Code gets messy
  • Repetitions occur
  • No owner of the code remains
  • Becomes unmanageable

Solution approach

  • Break down the big project
  • Perform small services
  • Manage each service with separate AI context

In this way:

  • Context remains small
  • It becomes intervenable
  • The code is not corrupted

If the spine is not intact, this approach will not work either.

b) Breaking existing clean project with AI

The most dangerous scenario.

In particular:

  • If there is a custom framework
  • If there is an internal library

AI:

  • He doesn’t know your nature.
  • Uses old version information
  • Generates deprecated config
  • Adds code repetitions

After a while:

You become unable to recognize your own code.

Critical risk

If you are not familiar with the code:

AI tells you “I did it” but:

  • He may have written raw query instead of ORM
  • It may have left a security vulnerability
  • It may have impaired performance

And you don’t realize it.

At this point:

You, not the AI, become the risk.

2. Research

The risk is greater here.

AI:

  • Can give old information
  • May give incorrect information
  • Can adapt

If you don’t verify:

In real life:

  • You make the wrong defense
  • You clash with experienced people
  • You lose confidence

Critical error

AI:

Using it as a “responsive system”

Correct:

“Using it like a search engine”

So:

  • Inquire
  • Compare
  • Search for resources
  • Ask for evidence

Real world problem

AI tells you:

  • “This is how you can do it”

But:

  • Service does not work
  • Data is not correct
  • The system is at POC level

You build a product on top of it → it explodes.

3. Entertainment

No problem here :)
In fact, it is the healthiest area of use.

Result

AI, if you don’t set limits:

  • Makes his own decisions
  • Builds architecture for you
  • Produces things you don’t know

And often:

Does not match your needs

The clearest takeaway

If:

  • If you don’t know what you’re doing
  • If you are not in control

You’re not working with AI, AI is using you.

Last word

If you are not at the table:

You are not part of the game, you are in the gamble.

Reference

My previous thought post that complements this article: Customer or Operator

  • ai
  • software
  • engineering
  • architecture
  • productivity