Junior's Basics: What Is Actually Under the App?

Layer 0: what you see

You drop a video file somewhere. A little later, you get back a transcript, the interesting scenes, and a draft title, description and social posts.

That’s the surface. Everything below it is what this post is about.

Layer 1: a place to put things (storage)

The video lands in a bucket: storage in the cloud, a bit like a folder on a computer that lives in someone else’s data centre and never runs out of space.

Cloud, the first time it appears: using someone else’s computers over the internet, paying for what you use, instead of buying and running your own.

Layer 2: something notices (an event)

When the file arrives, the storage service sends out a small message: “a new file is here.” Nobody is clicking a button. The arrival itself is the trigger.

This is called event-driven: things happen because something else happened.

Layer 3: a list of steps (a workflow)

The message starts a workflow, a written-down list of steps with arrows between them: first do this, then that, and if this fails, try again. On AWS I use Step Functions for that. You can literally look at it as a picture.

Software, the first time it appears: a precise list of instructions a computer follows, step by step, every time, without getting bored.

Layer 4: the steps do one thing each

Each box in the workflow is one small job:

1. Turn speech into text. A transcription service listens to the audio and writes it down.
2. Look at the pictures. An image-analysis service finds scenes and what is in them.
3. Write something new. A language model reads the transcript and drafts titles, summaries and posts.
4. Save everything back into storage, so it can be reused.

AI, the first time it appears: software that learned patterns from huge amounts of examples, so it can do things nobody wrote exact rules for, like recognising speech or writing a summary.

Notice where AI shows up: in three boxes out of many. Most of the system is plain, boring software moving files and keeping track of steps. That’s normal, and it’s good. The boring parts are what make the clever parts reliable.

Layer 5: someone else’s computers

Under all of those services are real machines in real buildings. I never see them, never patch them, never replace a broken disk. That is what I am paying for.

And because I only pay when a video goes through, an idle week costs me almost nothing. (Senior would want me to tell you that this is exactly the trade-off from the hybrid basics post.)

You can see the pattern now

Almost every modern app you use is the same five moves:

1. Something arrives.
2. Something notices.
3. A list of steps starts.
4. Each step does one job, and a few of those jobs are “AI”.
5. All of it runs on computers you never see.