Viyat Jhaveri
← Back

Getting Into AI Dev Work

A starter guide for people who are curious about building things with AI and are not programmers. It is meant to get you from zero to “I built a small useful thing” in a weekend, not to teach you computer science.

Every few weeks someone asks me what I have actually been doing with AI, and whether they could do it too. This is the honest, no-jargon version of that answer, written down once so I can send it instead of repeating myself badly over drinks.

You do not need to be a programmer to start. The whole point of the current tools is that you mostly work in plain English and the AI does the fiddly parts. Read this top to bottom once, then just start on the “Do this first” section.

What “AI dev work” actually is

Forget the hype for a second. In practice it is one simple loop:

  1. You tell an AI assistant what you want, in normal words.
  2. It writes the files, code, or notes to do it.
  3. You look at what it did, approve or correct it.
  4. Repeat.

That is it. The skill is not memorizing syntax — it is learning to describe what you want clearly, keep your stuff organized so the AI can understand it, and review its work instead of blindly trusting it. Those are judgment skills, not coding skills, and you already have them.

The magic is not that the AI is smart. It is that it can read and write your actual files and take real actions, so it becomes a tireless assistant instead of just a chatbot.

The one mental model that matters

Think of the AI as a very fast, very literal junior assistant who has read everything but knows nothing about your life until you tell it.

So the game is: put your context into files the assistant can read, and let the files be the source of truth — not the chat. Chats disappear. Files do not. If the important stuff lives in organized text files, you (and any AI) can pick up exactly where you left off, days later, with zero re-explaining.

If you remember nothing else: the files are the brain. The AI is just the hands.

The tools to start with

Do not overthink this. Pick one and go.

My honest recommendation: play in a normal chat app for a day to get comfortable, then install Claude Code and do one tiny real project. That is the jump that made it click for me.

Do this first: a weekend-sized project

Do not try to build something impressive. Build something tiny and real that you will actually use. Below is a menu grouped by what you are in the mood for — pick exactly one and stop reading the list. The specific project barely matters; they all teach the same loop. If none jump out, do the first one under “If you like organizing” — it is the gentlest start.

If you like organizing things

If your computer is a mess

If you write a lot

If you plan things for other people

The goal is just to feel the loop: you describe → it acts → you review. Once that clicks, everything else is variations on it.

Two rules that will save you pain:


The worked example: PersonalOS

The thing I have been building is what I call PersonalOS — not a fancy app, just a folder of plain text files that runs my life outside work: projects, travel, health, finances, coursework, home stuff. When I open my AI assistant, it already knows what is going on in my life and can help immediately — plan my week, triage my inbox, prep for a trip — without me re-explaining anything. It started as one file and grew only when the manual version got annoying.

That is the payoff of everything above: once your context lives in organized files, an AI assistant stops being a toy and becomes genuinely useful every day.

When you are ready to build your own, I have a full starter kit — the reasoning behind each file, what goes in it, and a “set it up this weekend” walkthrough. Email me at vjhaveri@me.com and I will send it over. Do not dive into it yet, though — do the tiny project first. The kit makes way more sense once you have felt the loop.

If you would rather just have the files, the skeleton is on GitHub: github.com/ViJhav/personalos-starter. It is empty by design — every file is a template, and the included CLAUDE.md interviews you to fill them in. Download it, point Claude at the folder, and say “let’s set this up.”


The handful of principles that actually matter

Everything else is detail. These are the load-bearing ideas:

  1. Files are the source of truth, not the chat. Organize your stuff so an AI can read it.
  2. Describe clearly, review honestly. Your job is good instructions and good judgment, not typing code.
  3. Start small and manual. Automate only the boring parts, only after you trust them.
  4. Keep it boring and simple. The best system is the one you will still use when you are tired. No dashboards, no over-engineering.

How to actually start this week

  1. Spend an evening in ChatGPT or Claude just using it for a real task — planning something, organizing something. Notice where it is great and where it forgets things.
  2. Install Claude Code. It is about a ten-minute setup.
  3. Pick one tiny project from the list above and do it.
  4. Once it clicks, look at the next layer — connecting the AI to real tools (calendar, email, your notes) so it can actually do things, not just write about them. That is where it gets addictive.

Seriously, the barrier to entry is way lower than it looks. The people who get good at this are not better coders. They are just the ones who started.

If you get stuck, or you want the PersonalOS starter kit, write to me at vjhaveri@me.com. I read everything, and I am always glad to hear what someone built.

— Viyat