Trust Issues with My AI Rubber Duck

As a neurodivergent dev, I genuinely love AI tools.
They help me start when starting feels impossible.
They hold context when my brain’s buffering.
They explain things without judgment — and they don’t care how many times I ask the same question.

For brains like mine — scattered, creative, deeply contextual — AI is like a second brain that doesn’t get tired.
Or distracted by snacks.
Or stuck trying to remember what tab it was in.

But it’s not magic.
It doesn’t know what you’re building.
It doesn’t care if the code works.
And if you can’t spot when it’s wrong, you’re not coding — you’re just vibing.
And that’s risky. Especially when you’re still learning.

Because sometimes, you ask it for help…

And it says:
"It should be fine."
Then gives you nothing else.

Should. Be. Fine.
The most dangerous phrase in tech.

So in this issue, let’s talk about how to actually use AI well:
Not as a shortcut.
Not as a replacement.
But as a tool you can drive — with clarity, curiosity, and control.

You don’t have to be perfect.
But you do have to be present.

AI is a gift for the scattered brain.
It turns vague intentions into rough drafts.
It remembers what you said five prompts ago, even when you don’t.
It turns "I know what I mean but I can’t word it" into structured code, explanations, or plans.

For neurodivergent devs, that’s gold.

Because when executive function bails, when task inertia hits, when your brain is spinning but not shipping, AI can nudge you back into motion.

It can scaffold ideas.
Translate chaos into steps.
Be the patient, predictable, always-available rubber duck that never judges or forgets.

It doesn’t even blink when you say, "Explain this like I’m five, but with more TypeScript."

And when it works? It really works.
Like "I solved a bug, wrote a test, and answered that one-week-old Slack message" kind of works.

But here is where it doesn’t

AI doesn’t know what you’re actually trying to do.
It doesn’t understand your system.
It doesn’t know that you renamed a function twelve times and lost the plot halfway through.

It just guesses.

And when you’re tired, overstimulated, or mid-context-switch from fixing a button to rebuilding the entire UI, it’s easy to miss the guessing part - because it sounds so confident.

It might:

• Make up a feature that doesn’t exist
• Suggest a method from an older version of the tool you’re not even using
• Chain together logic that technically runs, but also technically summons demons
• Add three layers of indirection and call it a performance boost

It doesn’t know what it broke.
And it definitely won’t admit when it’s just guessing with a straight face.

So you copy. You paste. You move on.
And suddenly, you’re looking at a file you don’t remember writing - because you didn’t.
It was a co-write with your emotionally unavailable AI co-pilot.

The vibe spiral

It starts innocently.

You paste in your error. AI gives you a suggestion.
You don’t fully get it, but it looks okay.
You run it. It works. You move on.

Later, something breaks.
You try to trace it back.
But now it’s 1 AM, your brain has 4% battery, and the code looks like it was written during a full moon.

You didn’t write the logic. You can’t explain the fix.
You just vibed it into your codebase — and now it’s your mess to mop up.

That’s not failure.
That’s not you being lazy or dumb.

That’s the price of too much trust and too little friction.
And we’ve all been there.

How to actually use AI (and stay in the driver's seat)

You don’t need to avoid AI.
You just need to stop letting it drive the car while you nap in the trunk.

Here’s what helps:

• Ask it to explain, not just output
  Don’t say "write this for me." Say "what is this doing and why?" or "explain this like I’m smart but also kind of melting."
• Don’t paste what you can’t talk through
  If you couldn’t explain it to a teammate (or your future self at 2 AM), it’s not ready to ship.
• Use it to learn, not hide
  AI won’t judge you. But that doesn’t mean you should disappear behind it.
  You’re not cheating — but you are in charge.
• Let it scaffold, then take over
  AI is the friend who builds the IKEA frame, but you’ve still got to screw in the shelves and double-check the instructions weren’t hallucinated.
• You are the dev
  Whether you're two weeks in or ten years deep, you’re still the one making decisions.
  AI is autocomplete with attitude. You’re the one with taste.

Final thoughts

AI isn’t replacing you.
But it can gently gaslight you into thinking you're being productive while you're actually just co-writing a fever dream.

For neurodivergent devs, AI can be a life-saver.
It can get us unstuck.
It can make the scary parts less scary.
It can hold our hand through the fog.

But it doesn’t care if you learn.
It doesn’t know if you're lost.
And it definitely won’t stop you from deploying garbage - it’ll just help you format it.

So yes, use it.
Lean on it.
Let it catch you when your brain has tabs open you can’t find anymore.

But don’t give up the wheel.
Because someday, it’ll give you a solution that looks perfect on the surface... and is completely wrong.
And when that day comes, you’ll know exactly why.
Because you were driving.

AI isn’t going anywhere and honestly, neither are we.
We’re just figuring out how to work with it, without getting ghosted by our own code.

If this issue helped you feel a little more seen, or made you laugh at a bug that gaslit you last week, consider hitting reply and telling me about it.
Or just whisper "it should be fine" to your terminal and hope for the best.
That works too.

I’ll be back next week - assuming I don’t forget to schedule the email again and end up hitting send at 6:43 AM while brushing my teeth.
Which... let’s be honest, is on brand.

See you soon,
Simen (the human)
Buffer (still loading)
Ping (emotionally over it)
Echo (watching you copy-paste that AI code without reading it)
Zap (enabled Copilot in Vim and won’t stop bragging)