Talking to an AI About What's on Your Mind

Programmers have a decades-old debugging technique that sounds like a joke: when your code breaks and you can’t see why, you explain it — line by line, out loud — to a rubber duck on your desk. Somewhere around line twelve, mid-explanation, you stop:“…wait.” You’ve found it. The duck said nothing. The duck never says anything. And yet the method works so reliably it has a name in every engineering textbook: rubber-duck debugging.

It works because explaining is not the same mental act as knowing. To explain a thing you must serialize it — put it in order, name its parts, commit to claims. Gaps that hide comfortably inside a vague understanding have nowhere to hide inside a sentence. The duck isn’t the mechanism. The articulation is the mechanism.

Now notice: the tangled situation at work, the decision you keep deferring, the mood you can’t name — these are also systems whose behaviour you don’t fully understand. They, too, respond to being explained out loud, line by line.

Talking to an AI about what’s on your mind is best understood as exactly this — rubber-duck thinking, upgraded. Everything that made the duck work is still there: you articulate, you hear yourself, you catch the “…wait.” What’s added is what the duck could never do:

• It asks the follow-up. “You said the meeting was ‘fine, mostly’ — what’s in the mostly?” The duck lets the evasion slide. A conversation doesn’t.
• It holds the thread. Twenty minutes in, it can point at something from minute three — the option you dismissed a little too fast, the name that keeps recurring. Out-loud thinking usually loses its own trail.
• It gives you the residue. Spoken insight famously evaporates by Thursday. A session that ends with a written summary converts the talking into a note you can read back — the exact combination the writing-versus-talking trade-off says you want: talk to discover, write to keep.

There’s also a quieter advantage worth naming plainly: a duck has no feelings to manage, and neither does this. No composing yourself, no watching the clock out of politeness, no editing the story so you come off reasonably. For thinking out loud, the absence of an audience to perform for is not a downgrade from human conversation. It’s a different instrument entirely.

You don’t know what you think until you say it aloud.

Being clear about the mechanism means being equally clear about its edges. An AI conversation is a thinking tool — it is not a person who knows you, and it is not professional care. If what you’re carrying is heavy — persistent, frightening, or bigger than a bad month — the right room is a licensed professional’s, and no amount of articulate ducking substitutes for it.

The tool is also only as honest as you are with it. A duck can’t catch you performing; an AI mostly can’t either. If you tell the flattering version of the story, you’ll get a well-organised flattering version back. The “…wait” moment only arrives for people willing to explain the thing as it actually is.

• One system per session. The engineer debugs one function at a time. Bring one topic — the decision, the conversation, the mood — not your whole stack. (This is half of the core reflection technique.)
• Explain, don’t summarise. The duck method fails if you skim. “Work is stressful” is a summary; “here’s what happened Tuesday, in order” is an explanation. Detail is where the “…wait” lives.
• Put an end on it. Open-ended talking drifts into rumination with company. A time-boxed session — with a visible clock and a written conclusion — keeps the exercise an exercise.

That’s the whole practice: one topic, explained properly, inside an edge, ending in a note. The duck would be proud — and would, as ever, say nothing.