It’s 10:19 PM on July 8th.
In exactly a week and a half, I’ll graduate.
Like most of my batchmates, I should probably be polishing my résumé, applying for jobs, or joining a full time role.
Instead...
I’ve spent the last few months obsessing over a question that refuses to leave my mind.
What actually happens after we press Enter?
still thinking…
I thought I knew.
I was wrong.
Every day...
Millions of us repeat this exact ritual hoping something unusual happens each time.
Yet on taking a closer look...
I had absolutely no idea what happened inside those mere two seconds.
…but
What if we could somehow FREEZE those 2 seconds...
Stretch them into 20 minutes...
Walk through every millisecond...
And meet every invisible system working before the first word even appears?
Let’s try to see through those 2 seconds!
Stop 1 - The City Gate
Imagine arriving at an airport.
You walk toward immigration.
The officer doesn’t ask
“Where are you going?”
The first question is much simpler.
“Who are you?”
Only after that answer is trusted…does the conversation even begin.
The internet works surprisingly similar way.
Before an AI can remember your previous conversations...
Before it can retrieve your documents...
Before it can even make it personal for you…
It has to answer one simple question.
Who just pressed Enter?
That tiny login screen we usually ignore is quietly solving one of the hardest problems on the internet.
Not intelligence.
Identity.
Imagine if ChatGPT never asked anyone to sign in.
Open your laptop → Ask a question → Now close it and open sometime later?
The conversation is gone.
Open another browser.
A completely different conversation..
Every visit feels like meeting a stranger.
No memory.
No personalization.
No ownership.
No trust.
Authentication isn’t there because engineers behind love login screens.
It’s there because the internet needs a reliable way to recognize you across time, devices, and sessions.
The moment you successfully sign in, something subtle happens.
The system now has a trusted identity to associate with every future request.
Suddenly...
Conversations persist.
Preferences follow you.
Memory becomes possible.
Personalisation begins.
All because the system finally knows ‘who is asking’.
From this moment onward, every prompt isn’t just a prompt.
It’s YOUR prompt. And computers care deeply about ownership
This tiny distinction we don’t see, unlocks everything else.
Stop 2 - Memory Isn’t Just for Humans
Did you ever stop reading halfway through a book...only to come back a week later and wonder,
“Wait... what was happening till now?”
Or maybe say : you’ve met someone a few times.
The 1st time, you ask,
“What’s your name?”
The 2nd time...
you’d hesitate a bit.
The 3rd time...
you remember it effortlessly perhaps.
But something interesting also happens.
You probably remember their name.
You might also remember what they do.
You might even remember where you met them.
But can you recall every single conversation you’ve ever had with them?
Probably not.
And that’s completely okay.
Because not every conversation deserves to be remembered forever.
Humans naturally decide what is important.
We remember names.
We forget small talk.
We remember birthdays.
We forget what someone ordered for lunch 3 weeks ago.
Our brains are constantly deciding what deserves a permanent place in there and what can quietly fade away.
Surprisingly...
Computers have to solve - this exact same problem.
Imagine telling ChatGPT:
“My name is Charan”
A few minutes later, you ask,
“What’s my name?”
How does it know the answer?
Did the model suddenly become conscious?
Not quite.
Something much simpler is happening.
Before the model even reads your latest message...
another invisible system quietly prepares everything it might need.
Your previous messages, i.e., earlier questions.
The model’s earlier responses (its answers).
Relevant memories.
System instructions.
Anything important enough to help answer your next question.
The prompt you typed...is no longer the prompt the model receives.
It’s been expanded.
Quietly.
Almost as if behind your back.
But here’s another fascinating part.
Not every piece of information is treated equally.
Your name?
That might be useful across many future conversations.
But the bug you’re debugging today?
Probably only useful during this conversation.
An instruction like
“Never explain with heavy mathematics.”
That might deserve to stay much longer.
Just like humans...
AI doesn’t throw every memory into one giant bucket.
It distributes them.
Some memories are short-lived.
Some stay with the conversation.
Some survive much longer.
This also explains something many of us have experienced.
Have you ever been deep into a conversation with Chat GPT or Claude?...
maybe 100s of messages in...
and suddenly it forgets something you mentioned at start of the convo?
It feels confusing and disappointing.
But it isn’t forgetfulness. It’s a trade-off. Just like us
It’s engineering.
Every model has a limited amount of context it can process at once.
As the conversation grows...older parts eventually have to make room for newer ones.
So TL;DR:
The next time you ask a follow-up question...
something remarkable happens.
You aren’t sending just one sentence.
You’re sending your latest message...
plus carefully selected pieces of everything that came before it.
And that quiet act of rebuilding context...
is one of the biggest reasons modern AI feels like a friend instead of a search engine.
Stop 3 - Finally... The Model
May be…this is something we’ve been thinking all along?...
Surprisingly...much later than most of us expected.
If this blog were a movie...the model would only appear in the final act.
This idea completely changed how I think about AI.
For years, I imagined the model doing almost everything.
In reality...
most of the work had already happened before it even saw my question.
Identity had already been established.
Memory had already been rebuilt.
Context had already been collected.
The city had already done its job.
Only now...
does the model finally get to thinking.
TL;DR(2):
Let’s say you’re a chef.
Someone walks into your restaurant and says,
“Make me something nice.”
You’d probably ask questions.
Do they like spicy food?
Are they vegetarian?
Do they have allergies?
The quality of your cooking depends on understanding what the customer actually wants.
AI models behave surprisingly similarly.
Before generating even the 1st word...
they try to understand the context surrounding every word you’ve written.
Not just the words themselves...
but how those words talk to one another.
Then...
something interesting happens.
The sentence you’ve written quietly becomes numbers.
Those numbers become mathematical relationships.
Those relationships become probabilities.
And from millions of possible next words...
the model starts making well calculated guesses.
One token/word at a time.
1000s of times.
Until a full answer appears on your screen.
Ironically...
the “thinking” we usually give all the credit to...
happens almost at the very end of the journey.
The 2 Seconds I Never Saw
Final TL;DR:
When I first started writing this article...
I thought those 2 seconds were mostly spent inside the model.
Now I see them differently.
They’re filled with 100s of tiny decisions quietly happening before the model ever gets a chance to think.
Authentication → Memory →Context → Retreival → Routing → Scheduling → Generation → Streaming
Each one almost invisible on its own.
Together...
they create the illusion of magic.
Maybe that’s what good engineering has always been.
Not making something complicated.
Making something complicated feel effortless.
The Last 30 Seconds..
It’s now well past midnight.
The same laptop is still open.
The same coffee mug is sitting beside me.
I’m also exactly one and a half weeks away from graduating.
When I began writing this blog...
I wanted to explain AI systems architecture.
Somewhere along the way...
I became sure I wasn’t aiming to begin with systems at all.
I wanted to spark genuine curiosity in others.
About slowing down moments we usually rush past.
About asking questions we’ve quietly accepted for years without ever stopping to wonder,
Why?
If this article made you pause...
even for a few moments...
and see those 2 seconds a little differently than before...
then I think it has done its job.
This is the first page of a notebook I've wanted to start for a long time.
I don’t know exactly where this blog goes next.
But I know the kinds of questions I want to ponder.
The invisible ones
The ordinary moments hiding extraordinary engineering.
If you’re as curious about those moments as I am...
I’d love for you to join me.
I already perhaps know the next question I want to explore.
Why does AI sometimes confidently answer something completely wrong?
Maybe that’s where we’ll slow time next.
Until then...
Signing off.
- memboard.winger (thats my username somewhere, leaving for you to figure out where:)
I love how you write