I started using AI the way most people do.
I had a task. I wanted it done faster. I typed a question. Something came back. I kept going.
At first, that felt like a small miracle.
Not because AI was perfect. Because it was there. Ready. Useful. A little bit like having someone smart sitting next to you while you work.
That part still gets me, if I am honest.
There is something genuinely exciting about asking a question and getting a useful answer in seconds. It feels like the future arrived early and decided to be helpful.
But after a few months, I stopped thinking of AI as a shortcut.
I started thinking of it as a relationship.
And like most relationships, it got more complicated once the novelty wore off.
The part nobody warns you about
AI is very good at sounding right.
That is the whole problem.
It can take a half-formed idea and return something that looks finished. Clean structure. Confident tone. Sentences that make you think, yes, that is basically what I meant.
Then you read it again.
And you realize you do not fully agree with your own idea yet.
That happened to me more than once. I would ask for a plan, get a beautiful plan back, and feel weirdly committed to something I had not actually thought through. The answer arrived before the thinking did.
That is when I started to get suspicious of the ease.
Not of AI itself.
Of my own confidence.
It stopped feeling like a tool and started feeling personal
When a calculator is wrong, you do not feel hurt.
When AI is wrong, sometimes you do.
That surprised me.
I think it is because we do not treat it like a calculator. We talk to it. We explain ourselves. We give it context. We start to expect it to understand us.
So when the response misses the point, it can feel less like a bug and more like a misunderstanding.
Like you were not clear enough.
Or it was not listening properly.
Or maybe both.
Either way, the emotional layer is real. We start by using AI. Then we trust it. Then we get annoyed when it lets us down. That is a very human sequence for something that is not human at all.
I had to learn how it thinks, not just how to use it
This took me longer than I expected.
I kept assuming that if I asked a better question, I would get a better answer.
True, but incomplete.
AI does not think the way people think. It is not sitting there understanding your intent in the deep sense. It is working through patterns, context, probability, structure. It is calculating toward something useful.
Useful is not the same as true. Useful is not the same as right. Useful is not the same as yours.
Once I understood that, my whole approach changed.
I stopped treating AI like an oracle.
I started treating it more like a very fast, very articulate collaborator who still needs managing.
That sounds less romantic.
It works better.
The castle mistake
I made this mistake early.
I would ask AI to build the whole thing at once.
The whole strategy. The whole document. The whole plan. The whole answer.
And it would.
That was the trap.
Because if the foundation was weak, the output could still look impressive. A weak idea in a strong wrapper is still a weak idea. It just takes longer to notice.
Now I go slower than the tool wants to go.
One block at a time. One decision at a time. One check at a time.
I give it grounding. I set boundaries. I ask it to explain itself. I ask what it might be missing. I ask it to push back. Sometimes I ask it to argue the opposite case, just to see if my thinking holds.
That is not because AI is bad at big tasks.
It is because I am bad at noticing when a big task is built on sand.
I started keeping notes like a person who did not fully trust their memory
This was another sign that I was taking it seriously.
For bigger pieces of work, I began saving prompts, tracking steps, writing little notes about what worked and what did not. Not because AI failed. Because the work got complex fast, and complexity needs something to hold it together.
That habit is honestly half the reason I ended up building Tansei: somewhere at the edge of the screen to keep the prompts, links, and half-finished pieces close while the work is still moving.
That felt almost old-fashioned.
Prompts in one place. Decisions in another. Fallbacks somewhere else.
Very un-futuristic.
Also very necessary.
The deeper the work, the more the details matter. Context. Sequence. Assumptions. Rules. What you said first. What you forgot to say. What you assumed was obvious.
AI will not always rescue you from your own vagueness.
Sometimes it will confidently build on top of it.
The question I keep coming back to
I still catch myself asking this:
Am I getting a good answer, or just a confident one?
That question has saved me more than any prompt trick.
Because AI can sound calm, intelligent, and sure of itself without being right. It can reflect your assumptions back to you in a cleaner voice. It can agree with the framing of the question without challenging the question itself.
So now I test more.
I ask for reasoning. I ask for sources when that matters. I ask it to explain where an idea came from. I ask it what it is unsure about. I ask it to tell me what I might be wrong about.
Not because I want to be difficult.
Because speed without checking is how smart people make dumb mistakes.
That includes me.
It changed my work, but not in the way I expected
I thought AI would mostly save time.
It does, sometimes.
What changed more was the kind of attention the work demands.
I question more. I edit more. I pause more. I notice when I am outsourcing judgment instead of outsourcing labor.
That was not what I expected when I first opened the tool.
I also did not expect to feel more respect for the people who have been doing this work without AI for years. Writing from nothing. Building from nothing. Researching carefully. Sitting with a problem long enough for the real shape of it to appear.
AI can help you move through those tasks faster.
It does not replace the thinking underneath them.
If anything, it made that clearer to me.
What I would tell someone just starting
Do not confuse fluency with understanding.
Do not confuse a polished answer with a finished thought.
Do not ask AI to solve the whole problem before you know what the problem actually is.
Start small. Build one piece at a time. Keep your hands on the work.
The real skill is not getting the answer first.
It is knowing when you are ready to ask.
I am still figuring that out myself. I do not have a perfect system. I do not think anyone does yet. But I know this much: the easier AI makes it to generate something, the more responsibility we have to decide whether that something is worth keeping.
That is not a reason to avoid AI.
It is a reason to use it like an adult.
With curiosity. With skepticism. With enough humility to go back and check.
Because the future is here.
It just still needs us in the loop.
↑ Back to top