code2026

Resources for learning to code with the help of AI in 2026

Last updated on

This is a resource for someone wanting to learn how to code (with the help of AI) from scratch. It has three parts, which you are invited to jump between in any order.

  1. To understand code, you need to understand a little about how computers and the internet work. This extra material is not usually included in coding classes, but it’s the wax-on-wax-off portion that will make you the Karate Kid.

  2. You need to learn a specific language, in this case python. After all, you’ll need to look at the code the agent writes for you to some degree, and there’s no getting around that. So you still need to “learn to code” the old-fashioned way, as far as we can tell. The goal is not just learning python, but understanding the abstract notional machine that python targets. It’s a model of how computer programs execute, and it carries over to other programming languages.

  3. You need to know what to ask the agent for. Therefore you need to know what kinds of software there are, such as web sites and apps. You need to look closely at the ones you use every day and see what screens they have, and how they organize navigation and information. You need to know when you need a database. You need to know about security and privacy. You need to know about testing (QA, quality assurance). You need to know about app stores.

Then you’ll be an expert!

Table of Contents

The Computer

Addition circuits

Computers add numbers together by putting two simple electrical circuits together. It’s a trick, to cause hardware to do something we can call “adding numbers.” I hope this can demystify a lot.

Additional resources about this idea

Computer organization

Curated resources that explain more about how computers work, building up hierarchically from the individual gates.

The apps that run our code

Code is either compiled into an executable, or interpreted live within a host executable (the latter being what python does). These compilers/interpreters are the apps that run our apps. They can grow very large and become IDEs (integrated development environments) if they also take on additional tasks like debugging, designing user interfaces, or managing multiple target platforms.

Programming languages and what they have in common

Languages themselves are objects of study. Some of them make it harder to write bugs, believe it or not! Python, sadly, is not one of those.

Python

Computer Science Circles: What you want to do is work through this site. They cover python in just the right way: with in-page visualizations of what the notional machine is doing.

I offer some supplementary material to look at alongside.

The notional machine

What you want to come away understanding is the following, which will allow you to understand every single programming language at once.

  1. Running programs live in memory, which is divided between a call stack (the “frames” in the CS Circles visualizer) and a heap (called “objects” in the CS Circles visualizer).

    See this for example in CS Circles section 10, reproduced here

  2. When code is executed, Python steps through the instructions, doing what each one tells it to in turn. That’s what the “Next” button simulates in the visualization above.

  3. Every piece of data is stored in a two-part structure. The first part says what type the data is, and the second part is the actual value.

  4. Memory for some types is allocated in the stack, and for others inside the heap. This shows up in the visualizer.

  5. Lists, dictionaries and other collections store references to other data rather than storing those values directly. They can be modified after they are created, e.g. a list can be extended. In the visualizer these are the arrows pointing out of the frame into the objects, as opposed to values like integers shown directly in the frame.

  6. When code is loaded into memory, Python converts it to a sequence of instructions that are stored like any other data. This is why it’s possible to assign functions to variables and pass them as parameters.

    Here’s an example of using functions as data

  7. Some instructions make Python read data, do calculations, and create new data. Other instructions control what instructions Python executes, which is how loops and conditionals work. Yet another instruction tells Python to call a function.

  8. When a function is called, Python pushes a new stack frame onto the call stack.

  9. Each stack frame stores variables’ names and references to data. Function parameters are just more variables in the frame.

  10. When a variable is used, Python looks for it in the top stack frame. If it isn’t there, it looks in the bottom (global) frame.

    Here’s a variant of an earlier visualization where inside add_something there is a local variable x in the frame, but also a global variable to_add that the interpreter accesses from the global frame

  11. When the function finishes, Python erases its stack frame and jumps backs to the instructions it was executing before the function call. If there isn’t a “before,” the program has finished.

That’s software in a nutshell! Here are some analogies about bits of the notional machine.

Notebooks

A notebook is another way to interact with Python (or R, or Mathematica, or others). It’s a document with three types of cells: formatted text, cells for entering code, and cells that show the output of the code. They usually are web pages. There’s a free site that offers notebooks called Jupyter everywhere. They run on your own comptuer in the browser, via a special version of python that’s also running in your browser.

Data types

Importing other modules, and connecting to the internet

Try this Jupyter Everywhere notebook which contains the following code to obtain a weather report. It imports a few modules: json, time, and urllib to help it do its job.

import json
import time
import urllib

def fetch(url, timeout=10):
    """Fetch a URL. Returns (status, headers, body-as-text).

    A request has to say who's asking. Many servers reject the default
    Python user-agent outright; an honest one that names the project is both
    politer and more reliable. Never lie about being a browser.
    """
    request = urllib.request.Request(
        url,
        headers={"User-Agent": "code2026-class-example/1.0 (teaching example)"},
    )
    # timeout is not optional in real code. Without it, a server that never
    # answers hangs your program forever.
    with urllib.request.urlopen(request, timeout=timeout) as response:
        return response.status, dict(response.headers), response.read().decode("utf-8")

weather_url = "https://api.open-meteo.com/v1/forecast?" + urllib.parse.urlencode({
    "latitude": 40.7128,      # New York
    "longitude": -74.0060,
    "current": "temperature_2m,wind_speed_10m",
    "temperature_unit": "fahrenheit",
})
try:
    # Pause between calls to a different service. One request is fine;
    # a loop making hundreds will get your address blocked.
    time.sleep(0.5)
    _, _, weather_body = fetch(weather_url)
    weather = json.loads(weather_body)
    now = weather["current"]
    units = weather["current_units"]
    print(f"  New York right now: {now['temperature_2m']}{units['temperature_2m']}, "
          f"wind {now['wind_speed_10m']} {units['wind_speed_10m']}")
    print(f"  (reading taken at {now['time']})")
except Exception as error:
    print(f"  weather lookup failed: {type(error).__name__}: {error}")

Those modules come from somewhere. In this case the Jupyter Everywhere system has them installed because they are very common modules everyone needs. But if a given system doesn’t have a module, you can install it with pip install json or similar. Here are some links that provide context on this infrastructure:

Games (and game state trees)

In case you want to think about writing games or AIs that play games.

Security and privacy

Here I only wanted to share some stories, which will help you know what to do by counterexample.

  • Do not store secrets, i.e. your, the developer’s, own passwords and access keys (or, even scarier, those of your employer).
    • Uber, 2014. An engineer put an Amazon web services (AWS) access key into code he published in a public repository. Someone used it to download a file with 100,000+ drivers’ names and licence numbers. Sources: FTC revised complaint (PDF), FTC analysis of proposed order.
    • A solo developer, 2015. Pushed AWS keys to GitHub, noticed, deleted them about five minutes later. A bot had already taken them and started ~140 EC2 instances mining Bitcoin, and the dev was charged $2,375. Sources: The Register, Slashdot discussion.
  • Having users log in to your site is a big responsibility.
  • Storing images is a big responsibility.
    • Tea, July 2025: 72,000 images including 13,000 selfies and photo IDs. An app for women’s dating safety required a selfie plus a government ID. The images were stored in a cloud storage bucket with no authentication; anyone with the URL could list and download everything. Sources: American Bar Association technical and legal analysis, Engadget.
  • Using third party libraries in your web app is a commitment to applying vulnerability patches forever.
    • Equifax, 2017: 147.9 million people. An Apache Struts security patch was published on March 7. An internal notice went out on March 9. A scan on March 15 failed to find the affected system. Attackers got in in May and were not noticed until July 29. The fix was only to run pip install pip-audit && pip-audit. Sources: GAO-18-559 (PDF), House Oversight report (PDF).

Software and the prompting of

Front end, back end

The basic dichotomies and building blocks, mostly written specifically for people coding with AI.

Information architecture: how to present information

Look at the world of software around you and how it’s organized. Then you can prompt for those good patterns.

  • Mobbin flow: see many, many screens from apps. Be an observer of apps and sites you like, and bring that organization to your own.
  • How to Make Sense of Any Mess (Abby Covert): information architecture: “the way we arrange the parts of something to make it understandable.”

Databases

If you need to store data, you probably need a database.

git

Software is stored in git because it lets you go backwards and forwards in the change history, with metadata about who changed what, and with the ability to mark certain states with release tags. It also lets people go off on a new branch while others fix bugs on the old branch. It fits the software engineering discipline.

Software engineering

Specs

It remains to be seen if we need to retain this paradigm exactly, but everything here needs to reappear somewhere in the prompt-based development cycle, even if it’s in your head. Learn from your elders!

Building apps

Coding for devices has its own languages and UI idioms that are designed more centrally by one company, usually. And there are extra steps to build apps for an app store such as Apple’s.

Building web sites

Web sites can be built in many different ways, though there’s usually a mainstream way to do it at any given moment. If you launch something publicly, it needs hosting and databases and security.

Building terminal programs

Or you can build a little text-based app, old-school.

AI for coding

You will help write the book on this aspect. For now there are some good early discoveries about how to do this well.

QA

This is your job now. The agent can build software but you have test it and find bugs and crashes. (That said, there are skills to help the agent test in a real browser or device simulator and find problems on its own.)