Decode the technical questions that stop you

The plan, as the tool prints it

An AI coding tool is a program that takes your idea in plain words and writes the files of a project for you. A plan is the list of choices the tool prints before it writes anything. It ends with one question. Here is the running-club plan, close to what a current tool prints.

Plan: running-club site

Build with:  JavaScript, using the Next.js framework
Host on:     Vercel
Files to create:
  app/page.js          the home page
  app/trails/page.js   post and browse trails
  app/rsvp/page.js     RSVP for the weekend run
  app/tidy/page.js     paste a messy message, get a clean run post
                       (uses an AI model)
  data/club.json       a plain text file that stores the trails and RSVPs
Also create:  a .env file to hold secrets (the pass for the AI service)

Approve this plan?

You read it twice and half the names mean nothing. It feels like a test on vocabulary you have never needed, so maybe you close the window, or maybe you type yes and feel like a passenger in your own project. Your ability was never the problem; the unfamiliar words were.

When an AI coding tool prints a plan you cannot follow, what stops you is the terminology of software tools, not your ability. A term takes minutes to learn.

Read it like a contractor's quote

Read the plan again the way you would read a contractor's quote. You will never pour the foundation. You can still approve the quote, question the odd line, and notice when it drifts from what you asked for. The card below matches each line to the decision it carries.

Build with names two things. JavaScript is the language the site is written in. Next.js is a framework, and a framework is a ready-made structure the tool builds your site inside instead of starting from a blank page. Every file the tool writes is built inside that framework, so it is the one line on this card you cannot change cheaply later. Languages and frameworks: approve the pick decodes this line in full.

Host on names where the finished site will sit. A host is a computer that stays on and connected to the internet, rented from a company like Vercel. That is what lets other people visit the site. This line becomes concrete in Where software lives and in Hosting, renting a computer on the internet.

Files to create lists one text file per page, and the tidy page is the one that uses an AI model. The last file, data/club.json, is a plain text file that stores the trails and RSVPs. The endings .js and .json only say what kind of text file each one is.

Also create: a .env file names a locked drawer. Secrets are private values the site uses behind the scenes, and visitors must never see them. The one in this plan is the pass the site shows to the company whose AI service tidies the messages, the way a membership card gets you into a gym. Environments: why it works on your machine and breaks online shows where that drawer lives.

Approve this plan? The tool did not quiz you line by line; it asked for one signature, and a choice the tool makes without asking you is called a default. Before you sign, start a plain note in any notes app and call it your idea file: the note you add a line to in every drill ahead. Write one line, as short as "host is Vercel, the tool's default, accepted for now". Written down, a yes you could not explain becomes a decision you can revisit once you know the words.

Read the tool's plan the way you read a contractor's quote: you approve it, question the odd line, and notice when it drifts from what you asked for. You are not being asked to write the code.

Every question in the plan unwraps the same way

During a build, a tool sometimes stops to check whether a default suits you. You see the question in its barest form, "Vercel or Cloudflare?", with two names you do not know.

Set the names aside and look for the decision underneath. Almost every technical question comes down to three plain decisions: where something runs (which computer does the work), where information is kept, and who is allowed to do what.

Every technical question a tool asks is a small decision wrapped in unfamiliar names. It is almost always about where something runs, where information is kept, or who is allowed to do what.

You unwrap the question with the same three moves every time.

• Ask what it is really asking. Put the product names aside and name the decision: for "Vercel or Cloudflare?", it is where the finished site will live.
• Ask why it is smaller than it sounds. The famous options suit almost everyone, and while the site is small, most decisions can be changed later at a cost you can afford.
• Make the move. Take the default, write down what you took in your idea file, and keep building.

Run the same moves on a question tools sometimes ask about that data file: should this be a file or a database, a program built for storing and searching information? It is really asking how the site should remember things between visits, and either answer works for a first site. The choice is covered later, in Data, where information lives.

The obvious objection: the AI can explain every word

You could paste any line of the plan into a second chat and get an explanation in seconds. As a substitute for a map of your own, the second chat breaks down in two places.

You cannot judge an explanation you have no map for. An explanation of Vercel only helps if you already know what a host is and why a site needs one.

You cannot tell a flavor question from a money question. A flavor question is one where either answer works; a money question is one where the wrong answer costs money or trust. Telling them apart takes a working picture of the whole system, and you build that picture across this part.

What this part gives you

In this part you build the working picture that makes any plan readable. You learn:

• what software is made of;
• where it runs;
• how a tap on a button becomes an answer;
• what the language and framework lines mean;
• how to undo an AI edit;
• why a site that works on your laptop can break online.

Try it now

This drill, a short exercise, takes about ten minutes. The first path costs nothing. On the second, the tool bills a small amount for each answer it produces, so expect a few cents.

No setup: Open any AI chat you already use and paste one line: "I want to build [your idea]. Before you write anything, show me the plan you would propose, with the language, the host, the files, and one short reason for each." A chat cannot build it, but it will print a plan much like the one above. Pick the line that intimidates you most and run the three moves on it, on paper.

With your tools: Give the same idea to Claude Code, the AI coding tool we use in this course. Read the plan it proposes, then approve one default you can explain back. After your yes it creates the files in a new folder on your computer, and the next chapter opens that folder. In Codex or Cursor, two other coding tools, paste the idea, read the plan, and approve the default you can explain. If nothing is installed yet, The Setup Clinic walks you through installing one.

Either way, leave with the idea file. Give your idea file four lines:

• your idea, in one sentence;
• the default you approved, or on the no-setup path the default you would take ("host is Vercel, the tool's default, accepted for now");
• the line that scared you most, and what it decides;
• today's date.

Chapter Summary

• An AI coding tool's first answer to your idea is usually a full plan, naming a language, a host, the files it will create, and a line about secrets.
• Freezing at that plan is a terminology gap, not an ability gap, and each term takes minutes to learn.
• Read the plan like a contractor's quote: approve it, question the odd line, and notice when it drifts from what you asked for. Your job is reading the proposal, not writing the code.
• Every technical question inside a plan is a small decision wrapped in unfamiliar names, nearly always about where something runs, where information is kept, or who is allowed to do what.
• Unwrap any such question the same way: ask what it is really asking, ask why it is smaller than it sounds, then take the default and write down what you took.
• A second chat can explain any word, but you cannot judge an explanation without a map or tell a flavor question from a money question, and you build that map in this part.
• Stop and decode fully, instead of approving on autopilot, whenever a line touches money, passwords, secrets, or your members' data.
• Your artifact from this chapter is the idea file: your idea, the first decision you approved, the line that scared you, and the date.
• Next, What software actually is: text files a computer runs opens the folder a tool makes after you approve a plan.

Sources

• Vercel documentation (last verified July 2026).
• Cloudflare developer documentation (last verified July 2026).
• Next.js documentation (last verified July 2026).
• Anthropic Claude Code documentation (last verified July 2026).