HOW TO START CODING WITH AI WHEN YOU DON’T KNOW WHERE TO START

It’s time to update my 2023 post called How to start coding when you don’t know where to start. There are still a lot of relevant items, which I’ve included below, but WOW, things have changed.

This is – by far – not a complete list. It’s just things I’ve found and worked with that are useful for beginners.

I’ll do my best to keep this list up-to-date as things evolve, and I’ve probabaly missed a few things, so let me know in the comments if somethign should be included!

As before, I’d like to help under-represented folks who are interested in a similar journey, but do not know where to start. My LinkedIn profile is here, feel free to connect with me and reach out to talk, chat, whatever. I am here to help you get started and beyond. Please mention this post when connecting.

FIRST: USE AI TO GET STARTED, NOT TO SKIP LEARNING

With the help of AI, you do not have to start alone with a blank editor and a search engine. You can start with AI as a study partner, coding assistant, debugger, reviewer, and occasionally a patient rubber duck that does not mind when you ask the same question five different ways.

AI is a great way to get started coding. It is also a great way to fool yourself into thinking you understand code that you do not understand yet.

So here is my first bit of advice: use AI like a tutor, not like a vending machine.

Bad prompt:

“Build me an app.”

Better prompt:

“I am new to coding. Help me build a very small app one step at a time. Explain what each file does, wait for me to try each step, and ask me questions to make sure I understand.”

That is a very different experience. You are still doing the work. You are just getting help sooner, and that can make a huge difference.

Here are a few prompts I would use if I were starting today:

“Explain this error message in plain language.”
“Give me the smallest possible fix, then explain why it works.”
“Before writing code, tell me the plan.”
“Ask me three questions before choosing a programming language.”
“Review my code and tell me one thing I should learn next.”
“Write a test for this function, then explain what the test proves.”

The best thing about AI for beginners is not that it can write code. The best thing is that it can explain, re-explain, simplify, compare, quiz, and keep you moving when you are stuck.

PICK A TINY FIRST PROJECT

Do not start with a giant app. Start with something small enough that you can finish it.

Tiny is good. Tiny gets finished.

Here are a few good starter projects:

A personal homepage with HTML and CSS.
A Python script that renames files in a folder.
A JavaScript page that calls a public API and displays the result.
A small budget tracker that reads a CSV file.
A book list app that lets you add and remove titles.
A learning journal where you write down what you learned each day.

Then ask AI to help you break the work into steps. For example:

“I want to build a learning journal app. I am new to coding. Give me one small step at a time, and after each step tell me what concept I just learned.”

That is a pretty good way to learn.

TRY A FEW AI CODING TOOLS

There are a lot of tools now. Do not try to master all of them at once. Pick one, try it, build something small, then try another one when you are ready.

GITHUB COPILOT IN VS CODE

If you are already using Visual Studio Code, GitHub Copilot is one of the easiest places to start. It works right where you write code, which is important. You can ask questions about the file you have open, get suggestions as you type, ask for tests, and get help understanding errors.

The GitHub Blog is a good place to follow what is changing. A few useful posts and resources:

My beginner tip: do not just accept Copilot suggestions. Ask Copilot to explain them. Ask what could go wrong. Ask for a test. Ask for a smaller version.

CLAUDE CODE, WORKFLOWS, AND ROUTINES

Claude is doing some really interesting work around coding workflows. The Claude blog has a lot of good material, especially the Claude Code category and the Agents category.

If you want to understand how AI coding tools are becoming real developer tools, check out:

I like the idea of routines for beginners. A routine is just a repeatable way to use AI.

Try these:

Explain routine: Paste code and ask, “Explain this like I am new to programming.”
Debug routine: Paste an error and ask, “What are the three most likely causes?”
Review routine: Ask, “What is one bug, one style issue, and one learning opportunity in this code?”
Test routine: Ask, “Write one test for the happy path and one for a failure case.”
Journal routine: Ask, “Summarize what I learned today and suggest what I should try next.”

That last one sounds simple, but it is powerful. Keeping track of what you learned helps you keep going.

OPENAI CODEX

OpenAI Codex is another important tool to watch. The OpenAI Developers Codex blog has practical posts on longer coding tasks and agent workflows.

Good places to start:

Again, if you are new, start small. Ask Codex or any AI coding agent to help with one small task. Add a form. Fix one bug. Write one test. Explain one file. Do not ask it to build a giant project that you cannot understand yet.

MICROSOFT FOUNDRY, WORK IQ, AND AGENTS

If you want to move from using AI tools to building AI apps and agents, check out Microsoft Foundry. The Foundry Blog has a lot of current posts from Build 2026 that show where AI development is going:

The key ideas to watch: models, tools, retrieval, memory, evaluation, observability, and governance. Those words may sound big, but they all point to the same question: how do we make AI useful and trustworthy in real applications?

Microsoft is also talking about Work IQ, a new intelligence layer for Microsoft 365 that understands how work gets done across an organization. That is worth watching alongside Copilot Cowork and Microsoft Scout. AI is moving from chat to getting work done.

TRY MODELS FOR FREE OR CHEAP BEFORE YOU COMMIT

One of the best changes since my first post is how easy it is to try models. You do not need to start with a giant cloud bill or a complicated setup. There are free and easy ways to experiment.

GitHub Models

GitHub Models is a great place to start because it is built into GitHub, has a playground, and is designed for quick personal setup. GitHub describes it as free to use, with seamless model switching, which is perfect for beginners.

What to try first: open the model playground, pick a small model, ask it to explain a simple code sample, then switch models and ask the same question. You will learn quickly that different models have different strengths.

NVIDIA build

build.nvidia.com is useful if you want to try serious models and AI application patterns without starting from scratch. The site highlights free inference endpoints, model APIs, agentic skills, and blueprints.

What to try first: use a free inference endpoint or browse the blueprints. You do not need to understand every GPU detail to learn from the examples. Just look at how the pieces fit together.

Hugging Face Models

Hugging Face Models is one of the best places to explore open models. There are millions of models, and Hugging Face makes it easy to filter by task, popularity, license, and whether a model can run with hosted inference providers.

What to try first: search for a small text generation model, a coding model, or an embedding model. Read the model card. Model cards are great beginner material because they usually explain what the model is for, what data or license applies, and how to use it.

Also check out Hugging Face Spaces once you are ready. Spaces are an easy way to try demos and see what people are building.

Ollama

Ollama is one of the easiest ways to run open models locally. This is especially interesting if you want to experiment without sending everything to a cloud service. Ollama also has cloud options, but the local experience is the big win for learners.

What to try first: install Ollama, run a small model, and ask it to explain a simple program. Then disconnect from the internet and try again. Local AI is not always the fastest option, depending on your machine, but it is a great way to understand what open models can do.

My advice: try all four at some point. Use GitHub Models for quick model comparison, NVIDIA build for powerful hosted examples and blueprints, Hugging Face for the open model universe, and Ollama for local experiments.

WHAT SHOULD I LEARN FIRST?

OK, so where should you actually start?

Start with a project, then pick the language that fits.

If you want to build websites, start with HTML, CSS, and JavaScript.
If you want automation, data, or AI experiments, start with Python.
If you want Microsoft stack, cloud apps, or enterprise development, check out C# and .NET.
If you want to work with data, learn SQL sooner than later.
No matter what path you choose, learn Git.

You can ask people what language to choose, and the more people you ask, the more answers you will get. I still like data, so check out the Stack Overflow Developer Survey and GitHub Octoverse to see what developers are using.

THESE THINGS ARE STILL RELEVANT

Some of the links from my original post are still great, with a few updates.

GitHub

GitHub is still where you will likely store your code and connect with open source communities. Learn repositories, issues, pull requests, README files, and commits early.

Stack Overflow

Stack Overflow is still useful. AI can answer a lot of questions, but Stack Overflow gives you answers that have been discussed, challenged, edited, and voted on.

LinkedIn

LinkedIn still matters for career connections. Do not worry if your profile is empty at first. Start connecting, follow people doing work you are interested in, and share what you are learning.

Visual Studio Code

Visual Studio Code is still a great editor. It is free, runs on Windows, macOS, and Linux, and works well with GitHub Copilot and tons of extensions.

GitHub Codespaces

GitHub Codespaces is the cloud development environment I would point beginners to now. If you do not have regular access to a powerful computer, this can make a big difference.

Microsoft Learn

Microsoft Learn is still a great free resource for Azure, GitHub, AI, .NET, security, data, and more.

GitHub Skills

GitHub Skills is a great hands-on way to learn GitHub, GitHub Pages, Markdown, Actions, Codespaces, and Copilot.

freeCodeCamp

freeCodeCamp is still awesome for project-based learning.

Kaggle

Kaggle is still useful for datasets, notebooks, and competitions. If you are interested in data or AI, spend some time there.

Cloud free tiers

Cloud is still where a lot of real applications run. Start with Microsoft Azure, and also check out AWS, Google Cloud, and IBM Cloud if you want to compare.

Certifications

Certifications are still useful when they match your goals. For Microsoft, browse Microsoft Credentials. Good beginner options include AZ-900: Microsoft Azure Fundamentals and AI-900: Microsoft Azure AI Fundamentals.

A FEW SAFETY HABITS

Before you get too far, build these habits:

Do not paste secrets, passwords, private keys, customer data, or confidential work into AI tools.
Ask AI to explain assumptions.
Run the code before trusting it.
Ask for tests.
Read official docs when security, cloud, or money are involved.
Keep your projects small enough that you can understand every file.
Use Git so you can recover when experiments go sideways.

Here is a prompt I use a lot:

“Before you give me code, tell me what assumptions you are making. After you give me code, tell me how I should test it.”

That one prompt can save you a lot of time.

WHERE I WOULD START TODAY

If I were starting today, I would do this:

Create a GitHub account.
Install Visual Studio Code.
Try GitHub Copilot in VS Code.
Pick one tiny project.
Ask AI to break it into steps.
Build one step at a time.
Save the project on GitHub.
Try the same prompt in GitHub Models, Hugging Face, NVIDIA build, and Ollama.
Write down what each model did well and not so well.
Start the next tiny project.

You do not need the perfect plan. You do not need to know every tool. You do not need to understand every AI term before you begin.

You need curiosity, a small project, and enough patience to keep asking the next question.

The good news is that AI can help with that.

OH, AND DON’T FORGET…

Have fun!

brianbenz.com

Leave a comment Cancel reply