Learning to code takes time, so much time, but with the help of AI, you could make ideas into apps and websites with a mere prompt.
The year’s biggest buzzword and trend of AI clearly has no sign of stopping, affecting anything and everything in your life.
Often there to assist, artificial intelligence is being added to improve and change how things work, endowing devices and software with new features that can either seem like a waste or a superpower.
Phones with AI have image editing abilities and wallpaper creation skills, while some are given the more useful skill of translating languages in real-time using two screens.
Computers are taking advantage of AI with text creation seemingly from nothing more than a simple prompt or statement, as ChatGPT is written into Windows with CoPilot, and that’s not all.
You can make music from nothing by scripting your sound with words, and even have that language deliver a brand new realistic image out of the blue.
It seems there’s no stopping AI, so what’s next?
Code.
Autonomous programming
The idea of AI-enhanced coding isn’t new. Not by a long shot.
Ever since the promise of artificial intelligence started leaning its way into more software and solutions in the past few years, we’ve seen AI-enhanced code editors pop up.
The programming repository of GitHub was one of the first, offering an assistance for code competition and automatic programming, popping up in late 2021 just before the AI revolution really took centre stage.
Even before then, the concept of automatic code completion had been an idea the 50s, with subsequent work popping up in development environments over time.
Like most intelligence features, the idea is to make life easier for the worker, which in this case was the programmer. Automatically suggest and complete additions and code blocks, and the programmer can be focused on the more complex tasks and principles.
The path to a fifth generation language
The idea of talking and just having code execute has long been a dream of those in computing. Not just a dream, really, but an expected item in a chart. An eventual note that it will happen, so much that it’s been referred to as part of the evolution of computer language.
Before the age of true digital literacy in schools, I can recall sitting in my optional class of Computer Science and being told about the generations of programming languages.
First generation were the ones and zeroes, the binary we all know gets to the root of programming, while second-generation used assembly languages to make life a little easier than looking at sequences of two numbers representing on and off.
In school, I was learning third-generation languages. Basic, C++, C#, Pascal, and JavaScript are all common examples, and require a programmer to learn skills to understand how to build without needing to know binary or assembly. The concepts behind programming are often connected with a third-generation language, with the programming language being the next step.
Every language is complex and has its own rules and requirements, but as the generations evolve, so too does the ease of use to get stuck in.
Fourth-generation languages get human words and language into a spot where you still clearly need to know something about the language and code of programming, but life is made easier. Python is a big example, and you could likely class Apple’s SwiftUI in there as well, one of the ways kids get into learning to code with the likes of Swift Playgrounds.
Then there are fifth-generation programming languages.
Code in your words
Fifth-generation programming languages are the promise that your words and your language as you know them now will be able to program without any extra skills. They are the epitome of language as code: you tell a machine what you want and it makes it for you.
In many ways, it’s what artificial intelligence is perfectly suited for. While we see AI being used to enhance text or create media, the idea of a machine learning concept understanding great swathes of material to translate your words from one thing into another are exactly what AI is promised to do.
Essentially, AI programming is a fifth-generation language because it transpiles your words into another type of code. AI would be the converter and compiler for your actual words.
To do this, you simply need to find a system able to do just that. You need to find an AI counterpart that has been trained on codebases and software kits (SDKs) and programming interfaces (APIs) and all manner of documents and language to understand the difference between your words and the desired output, and then let it work out the difference.
Fortunately, there are already plenty to choose from, and the results deliver even initially.
Code in no time
AI is clearly having a year in 2024, and AI programming really pumps it up another notch. It’s staggering how well it works, and how it essentially undermines what is normally the development process for a programmer.
Developing something normally takes time. Lots of time. So much time, that the process is a combination between planning and learning and learning and planning, and often working with others to get your grand vision out in an executable.
It’s no wonder that learning to code is an undertaking with real rewards. You’re not just reading a book; you’re reading several and executing examples and lessons and results following those learnings. It is a process, and one that takes time.
Enter AI programming, which undermines all of that by having you enter a prompt and letting AI work it all out for you. Not only that, they can also run the code, building out programs from simple text prompts that can be small before growing and spiralling to much larger projects.
How to prompt an AI coding platform
There’s no shortage of the things, and more are popping up.
The clever AI system that is Anthropic’s Claude.AI can run small amounts of code in its browser, while Lovable’s AI generation system is a prompt to app system that can make compact sites and single page apps work well, too.
On the desktop, we’ve been getting acquainted with Codeium’s Windsurf, which does a stellar job of turning words into lines of code that you can incorporate into your apps, giving you a little more to work with than the AI-based autocomplete suggestions Apple’s current Xcode has working for it.
But our favourite has to be Stackblitz’s Bolt.new, a clever platform that literally sees your words turn into code, running that in your browser before giving you a chance to deploy it with ease.
The concept is so daft and ingenious that it’s a wonder it works as well as it does, spinning real language into real apps, provided you prompt it well.
Bolt does make things very easy, though, and even offers a way to fancy up your own words and prompt by making it easier for the system to understand. It’s not always perfect, but it can help a text prompt go from “meh” to “more useful” in seconds, getting a better outcome all along.
A real-world demo
Reading this as mere words wouldn’t be convincing enough, however; your brain needs to see action and results to understand the outcome, as well.
To do this, we spoke in real-world language to have AI build something we had in our kitchen: a Die Hard advent calendar of Hans Gruber falling from the Nakatomi Plaza in his final moments.
Every day in the lead up to December 25, we lower a little wooden Hans one level as he falls to his demise in the treasured Christmas classic Die Hard, a yearly favourite in this house. However, not everyone has the calendar, so asking an AI to make a digital version is a good test.
It’s a test that involves logic and calculations, taking into account how many days there are to Christmas, the date of the user’s computer, the position of Hans as he falls, and the problem of making it responsive and able to reflexively resize for the modern variables of mobile browsers.
Building a digital Die Hard advent calendar would be JavaScript and React and programming the likes of which my skills would not be able to match. Fortunately, all I had to say was:
“Make an advent calendar timer website that shows a man falling one level down a picture of a building every day before Christmas. Each day, the image of Hans Gruber falls one floor until the last day when he hits the ground, and the website says “Yippie ki yay. Merry Christmas.” For the rest of the year, show a countdown for how many days until Hans Gruber falls from Nakatomi Plaza.”
That was all Bolt needed to get to work, translating what my words roughly meant, and breaking the process down into core concepts and code generations, running it all in a layer for my browser immediately after.
It took a few other prompts to get the system working correctly, and I had to supply some artwork to bring the whole thing together, but the result works and you can see it for yourself.
What’s more, Bolt managed to deploy it for me, a process that might have taken several hours of prep time and delivery, and yet was whittled down to second by the service.
The outcome is a Single Page Application built from web languages that works — that really works — delivering the digital equivalent of my physical Die Hard advent calendar. It’s insane.
Understanding programming helps
Sending basic words to a text prompting system is a process anyone can try, but it’s one that can deliver unexpected results. You only have to look at how easy it is to misinterpret words on a screen to realise AI isn’t always going to nail every meaning from your words, unintended or otherwise.
Words can and will have multiple meanings, and so making sure your prompt is detailed and written in a way that’s descriptive and specific is important, but so too is knowing a bit of programming.
To make these prompts stand out, you may have to mention how you want databases to connect, or the exact functionality you want certain aspects to have. You might have to copy in errors and not be afraid of them, and you may need to be comfortable with getting your hands dirty and coding changes to things the AI coder just can’t do.
In using AI coding to help build a new social platform Standfirst, I relied on a combination of my own designs, prompts, and understanding of coding and databases to get a working, fully-realised proof of concept out that anyone could join and try. One or two prompts wasn’t all it required, but more skills than mere text on a screen.
If programming isn’t your forte, however, you might want to patch your ideas as a prompt to another AI system to get the right language in the first place. Anthropic’s Claude is one such approach, sending it a prompt such as:
I’m building the ideal prompt for an AI development platform to build my idea out. I need a well-structured, planned brief to pass on to help me build <insert the idea here>
A new world for creators
Even though my experiments show that you probably need more than a fistful of words to get an idea out of your head and onto the screen, that idea is crucial, and it could herald a new world for creators.
No longer hamstrung by the idea that building a project takes forever, creators with an idea can set out to build an “MVP”, the Minimum Viable Product needed before set out on the next step and look for funding. They can see if their idea has legs, whether it works in reality, and even try to find a market and audience for it without needing to hire developers.
It’s a whole new world, and it’s one that potentially opens it up for creators looking to make a difference with small beginnings.
