Joanah Fermin

Tutorials / Disciplined AI-Assisted Development · Lesson 7 of 14

On this page

    Explore: Talk It Through Before You Build

    Step 1 of 7.

    Step 1 of 7 1Explore 2Document 3Specify 4Build 5Test andAI review 6Humanreview 7Refine

    Before anything is written down or built, there is a conversation. You have an idea for a tool, and you do not yet know what it should do, who it is for, or how to build it. This lesson is about having that conversation with an AI, so that you end it with decisions and not just ideas.

    Our goal: know what to talk about, who brings what, and what to take away from an exploring chat. Everything here works in a plain chat window.

    Here is the route:

    1. What you are exploring, as a short list.
    2. Who brings what: the AI and you.
    3. Two sample chats about the bookshop back-office tool, one for each hat.
    4. What leaves the chat.
    5. When to stop.

    We leave out writing the project document (lesson 08) and writing a feature in detail (lesson 09).

    What to talk about

    Exploring has two halves. First you understand the problem. Then you choose how to solve it. Skipping the first half is how people build the wrong thing quickly.

    What to talk about 1. Understand the analyst hat The problem The goals Who it is for What is in, what is out 2. Choose the architect hat The options Good and bad of each Your decision 3. Write the decisions down

    The hats are from lesson 03. The analyst hat asks what the problem really is. The architect hat lays out the choices. One hat per pass keeps the answers clear.

    Who brings what

    Think of the AI as a technical head with a huge store of knowledge. It has learned from an enormous amount of software and writing about how many projects were built: what worked, what went wrong, and what each choice costs. So it knows the usual options and their trade-offs, and it can ask questions you did not think of. It does not know your situation, and it should not make your decisions.

    💡 A simplification: it is a very wide memory, not a perfect one. It can still have gaps and get facts wrong (lesson 02), so you check what matters.

    Who brings what The AI brings knowledge of many built projects the usual options their trade-offs questions you missed You bring facts about your situation what matters most the final decision The AI fills gaps. Only you can settle forks (lesson 02).

    Two chats, one for each hat

    Each hat gets its own short chat, so each one has a single job. These are examples written by hand to show the shape, and a real AI's answers will differ from run to run. The shop is made up.

    Two chats, two short lists Chat 1analyst Scopeshort list Chat 2architect Decisions+ questions paste it intoa new chat into step 2

    Chat 1: the analyst hat

    Understand the problem first. Start with the analyst hat and let the AI ask you:

    The analyst chat Act as a business analyst. I want a small back-office tool for my bookshop. Help me flesh out the features. What is your main pain point? What do you most want to fix? I have to enter about 2,000 books by hand, and it takes forever. If you have a barcode scanner, or the budget for one, the tool could scan each barcode and look up the book's details online. You would only check and save. Great! What is the summary of the specs so far? Here is the scope so far: - Login, because the tool is online - Scan a barcode to add a book - Search and list books to check stock That is enough scope for now. Blue: you. Pink: the AI.

    The AI brought an idea you had not thought of, scanning, because it has seen many tools like this (lesson 02). You decided it belonged in the scope. Asking for a summary gives you a short list to carry forward. If you are unsure how to begin, ask the prompt writer hat (lesson 03) to interview you first: "Help me write the request. Ask me what you need to know."

    Chat 2: the architect hat

    A new chat starts empty (lesson 02), so paste the scope in. Ask the AI to question you first, then choose:

    The architect chat Act as a software architect. The scope is: login, scan a barcode to add a book, and search and list books. Ask me what you need to know first. What languages do you know? Do you have somewhere to host it? C#, and basic HTML and JavaScript. No hosting yet, it is our first time. Budget is tight, about $20 a month. I suggest a C# web app with server-rendered pages (little JavaScript, which suits you), SQLite (free, one file) and a budget VPS. The alternative is managed hosting: easier, but it costs more. Which do you prefer? Your suggestion, with the VPS. Decided: C# web app, server-rendered pages, SQLite, hosted on a budget VPS. Still open: which VPS, and who looks after it? Blue: you. Pink: the AI.

    The AI asked about your situation before advising, and your answers (a tight budget, a first time hosting) shaped what it suggested. It gave a suggestion and an alternative, and you chose. It settled the direction (a budget VPS) but left the details open: which VPS to pick, and who looks after it? That is fine, because those details do not block the first features. Write them down as open questions and settle them before you deploy. Those two lists, the decisions and the open questions, are what you keep.

    Two terms from the chat. Server-rendered pages means the server builds each page, so there is little JavaScript to write. A VPS (virtual private server) is a rented slice of a server that you set up and look after yourself.

    What leaves the chat

    The chats themselves are not the product. Chats are long, they wander, and the next session starts empty (lesson 06). What matters is the short list at the end.

    What leaves the chat The chats long, wandering, forgotten next time Decisionsand the why Open questionsstill to settle Step 2the docs

    Keep the why with each decision ("SQLite: free, one file, enough for about 2,000 books"). A later session, or a teammate, can then see whether the reason still holds. And keep the open questions. An honest "we do not know yet" is better than a quiet guess, because a guess in the AI's hands becomes code.

    When to stop

    Exploring has no natural end, so decide one.

    When to stop Stop when the decisions fit on one page and a newcomer would follow Keep going if you cannot say what the problem is, or what to choose You will not settle everything. Open questions are fine.

    💡 Not too deep. Exploring decides the shape of the tool, not every detail. Details come when each feature is written out in lesson 09.

    Recap

    • Explore is step 1: a conversation that ends in decisions.
    • Two halves: understand the problem (analyst hat), then choose how to solve it (architect hat).
    • The AI brings the usual options and trade-offs. You bring your situation and the decision.
    • Use the habits from lesson 05: ask it to interview you, repeat your order, and give options.
    • What leaves the chat is a short list of decisions with reasons, and a list of open questions.
    • Stop when the decisions fit on a page and a newcomer would follow them.

    Questions to think about

    The answers are left out on purpose. Try them in your own words first.

    1. Why understand the problem before choosing a solution?
    2. What does the AI bring to an exploring chat, and what do you bring?
    3. Why keep the reason with each decision?
    4. Why keep open questions instead of letting the AI guess?
    5. Why is the chat itself not what you keep?

    Next lesson: step 2, Document. Turn the decisions into a project document the AI can read in every session.

    ← Overview of the ProcessDocument: Write It Down for Every Session →