In a roadmap review, an engineer says the new feature cannot work offline because the model runs server-side, and the room moves on while you are still translating the sentence. None of the words are hard, but there is no picture behind them, because software is a word most product people use all day without ever being shown what it refers to. In Decode the technical questions that stop you you practiced slowing those moments down and naming what is actually being asked. This chapter gives you the frame that makes the naming fast, and the frame is smaller than you expect.
Every piece of software is the same three parts
A piece of software is a set of instructions, a place where those instructions run, and an input that becomes an output.
Every product you have shipped, bought, or churned from is those three parts at some scale, and nearly every technical question you will ever face is a question about one of them.
Test the claim on things you touched today.
- A calculator. The instructions are a short list of arithmetic rules. They run on your phone. Tapped buttons go in, and a number on the screen comes out.
- Email. The instructions are far larger and run in two places at once, a little on your phone and mostly on computers your email provider owns. Your drafted message goes in, and that message appearing on the recipient's screen comes out.
- A maps app. The instructions compute routes. The heavy ones run on the company's computers while the display ones run in your hand. A destination goes in, and a route drawn over a map comes out.
The instructions can be a page of arithmetic or a model trained on a large share of the written internet, and the pattern never changes. Building something new means making the same three decisions deliberately: what the instructions do, where they run, and what goes in and out.
The three parts hold for an AI product too
AI products are where you would expect these three parts to break down, so let us walk one through them from start to finish. Granola is a meeting notes app. During a call it captures the meeting audio directly on your laptop, without sending a bot into the meeting, while you type rough notes into its pad. Afterwards, AI models running in the cloud take the transcript and your rough notes and produce polished, structured notes.
Hold that against the three parts.
- The instructions live in two places: an app on your laptop that handles the audio capture and the notes pad, and models on remote computers that generate the finished notes.
- The running happens partly on your laptop and partly in a data center you will never see.
- The input is meeting audio plus whatever you typed, and the output is a set of notes tidy enough to forward.
Laid out like that, you can point to exactly which part each technical-sounding question belongs to. No bot appears in the call because the capture instructions run on your laptop instead of dialing in as a participant. The app needs a connection because part of its instructions runs on computers it has to reach. When a privacy reviewer asks what the models receive, that is an input question with a precise answer, the transcript and your typed notes.
The same three parts hold for this site
The page you are reading is software too, and its breakdown is short.
- Instructions: a folder of plain text files holding the pages, the components, and the chapters themselves, every one readable in an ordinary text editor. We wrote this site with an AI coding agent. We described each page in English, the agent produced the files, and we read and corrected what came back.
- Runs: a rented host. We own no computer you could visit. When you opened this page, a machine in someone else's data center executed those files and sent back the result.
- Input and output: your tap on a link went in, and this chapter on your screen came out.
Nothing in that breakdown is exotic, and that is the point. If the frame holds a meeting transcriber and the page it is printed on, it will hold whatever you decide to build.
Every later chapter names one of the three parts
This frame is the map for the whole level, because each chapter ahead takes one part and goes a layer deeper.
- The instructions get their close-up in languages and frameworks, which covers what they are written in, and in Git, which keeps every AI edit reversible.
- The place they run is the business of hosting and of environments, where the mystery of code that works on one machine and breaks on another gets solved.
- The input and output run through the journey of a request, APIs, and data.
The move is available right now, though. When a technical sentence stops you, name which of the three parts it is about before you respond. Naming the part you are stuck on turns out to be most of the work of getting unstuck.
Try it now
No setup: Pick an app you used in the last hour and write its three parts in three lines: what the instructions do, where they run, and what goes in and out. For a podcast app the lines might read instructions that fetch, store, and play audio, running partly on your phone and mostly on the company's computers, with a tapped episode going in and sound coming out. Pick a boring app, because the boring ones prove the frame holds for everything.
With your tools: Open Claude Code and have it run the frame before any build. Pick a tiny idea you actually want, a bill splitter, a reading tracker, a form that collects feedback, and prompt it with: "Before writing any code, describe the three parts of this idea: what the instructions would do, where they would run, and what the input and output would be." Read the answer and check that you can restate each part in your own words, because the restating, not the building, is what this chapter is training. If Claude Code is not installed yet, the Setup Clinic gets you there in one sitting. In Codex or Cursor the move is the same: paste the prompt into the chat sidebar and ask for the three parts before any code gets written.
Chapter Summary
- A piece of software is three things: a set of instructions, a place where those instructions run, and an input that turns into an output.
- The instructions can be a page of arithmetic or a model trained on much of the written internet, and the three parts stay the same either way.
- The same three parts held for a calculator, an email system, a meeting notes app, and this page, so they will hold for whatever you build.
- A model is just instructions made by training on data instead of typed by a programmer, so run it through the same three questions, starting with where it runs.
- When a technical sentence stops you, name which of the three parts it is about before you answer, because naming the part is most of the work of getting unstuck.
- Building something new means making all three decisions yourself instead of inheriting them by default: what the instructions do, where they run, and what goes in and what comes out.
- Next, Where software lives opens up the part about where instructions run, the one this chapter could only point at.
Sources
- Granola product documentation on device audio capture and AI note generation.