Where does This Place AI-Assisted Development?
AI and the casual coder
I gave eight AIs the same web development task of creating a simple one-page website, using only HTML, CSS, and JavaScript. Each of them received the same prompt, stipulating they had to meet certain constraints, including: dynamic screen size adaption, international accessibility standards, and human-readable code. They also had to have certain content included: photos, testimonials, date and time details of an event, and a registration form. The scenario was fictional but based on real projects I’ve done.
Was it worth it? Back when I was creating volunteer websites on evenings and weekends, it would take me a couple of weeks to finish a web page. A lot of that was back-and-forth with the organizer, who’d want changes, or need to provide me with details they’d promised. Having a trustworthy AI tool available would cut that down to maybe a week, week and a half. Unfortunately, AI cannot help you get needed information from a volunteer coordinator with too many things to remember, so it can only help so much.
If anything, using AIs to make these sample web pages was a good reminder that a lot of coding work isn’t the actual coding. It’s getting the information you need to include and getting a version approved. I suspect a lot of the perception vs reality issues with AI boil down to people not taking into account where the actual time is being used up. Clients of all sorts can be famously vague about what’s needed until you start reviewing drafts with them.
Consider also that ethical standards hold the human responsible for AI output. That’s a compelling reason to take writing a detailed, thorough prompt seriously, and to be critical of what an AI will provide in response to a thorough prompt.
In short, while AI won’t help you with getting the content you need any faster, it can save you from late-night struggles with CSS, and free up time you would have spent getting a form field to behave.
Benefits outside of the computer
AI forces humans to stop and think about what they actually need and how to define it precisely.
“Vibe coding” — throwing casual prompts at AI and hoping for the best — is already being replaced by careful planning. This test was designed specifically to use a single prompt for consistency, but iterating can work well too, so long as you know where you need to end up.
No matter how carefully you work through your specs, you’re going to miss things. Collaboration is essential, and a good AI, like a good human team member, will question details and suggest improvements. AI has made excellent suggestions I’ve adopted gladly, even during this test. The point is: it works best as collaboration, not delegation.
Another piece of good news: AI can produce something genuinely useful for my imaginary frazzled volunteer. I’ve been that frazzled volunteer. I’m comfortable with HTML, reasonably fluent with CSS, and terrible with JavaScript (though working with AI is, ironically, forcing me to improve).
Judging the code against my past abilities when I did more volunteer work, any of the top test responses would have been enormously helpful for completing a real web page quickly.
Even without a preferred tool, I could do what I did for this test: work with AI to develop specs, edit them to ensure they matched real needs, feed them to several popular AIs, then continue with the best results. The initial evaluation took only a couple of hours — far less time than coding everything from scratch would have taken.
What’s the End Game?
People worry about AI taking jobs. It’s depressing, but helpful, to remember automation has been taking away jobs since the Industrial Revolution. This trend has already been happening for centuries.
But here’s the thing about any kind of automation: there’s always a trade-off. Knitting machines still can’t do things human knitters do with ease. Machine-made sewing pins are far less durable than hand-made ones. Mass-produced food lacks the nuance of meals cooked from scratch. People are complaining about AI slop, but the truth is they accept “slop” as normal in many other areas, like food, clothing, and furniture. That doesn’t make the slop good, but it points out that what’s considered “good” didn’t start sliding when AI became popular.
It’s Time to Power Up
Douglas Engelbart, who created the mouse, copy-and-paste between different apps, groupware, and many other things we take for granted as “how computers work,” always maintained that computers should augment human abilities. Augment, not supplant.
Total or near-total automation is rarely superior to human work. There will be trade-offs, because there always are. But just as people with knowledge, appreciation, and money will pay $1,000 or more for a hand-knit sweater rather than $5 for a machine-made one, people will continue paying for human-made, human-tailored writing and coding.
The sweater example reveals something else: in the future, there may be less direct connection between worth and monetary value. Most people with excellent hand-knit sweaters didn’t pay $1,000 for them. At most they paid for yarn — $30-$300. Then they did the knitting work themselves, or paid in kind through barter. I once received three jars of amazing homemade salsa, better than anything in stores, in exchange for designing and knitting a pair of socks. Both the salsa-maker and myself felt we got excellent deals, yet neither of us would have considered ourselves well-paid if cash had been directly involved.
Make or Buy?
So what does this test tell us about the core question: should you make or buy?
The answer depends on what you’re actually trying to accomplish.
If you need a simple website quickly and have some technical literacy, free AI can get you most of the way there. You’ll still need to verify accessibility, fix bugs and edge cases, and polish the details. But it may be better than starting from scratch.
If you need something complex, reliable, or mission-critical, AI isn’t ready to work unsupervised. You’ll need human expertise, either your own or someone you hire. The AI might help that expert work faster, but it won’t replace them. Yes, I know there are people claiming to have used AI to build entire applications, even though they’re a non-coder, but they all gave the AI a lot of freedom in how to solve the prompt they provided. The more you need specifics, the more AI can fall down, as shown in this test.
If you’re learning to code, AI is a double-edged sword. It can explain concepts and provide examples, but it can also create dependency. You might ship something functional without understanding how it works. That means you can’t maintain or improve it later.
The real skill isn’t choosing between making and buying. It’s knowing when to use AI as a collaborator, when to hire a human, and when to invest in learning the skill yourself.
Augmentation of any kind changes the value of human skill and knowledge. The trick is understanding how to work with it effectively, and appreciating that augmentation means we can do more, and therefore need to hold ourselves to better.
What’s your verdict? Has AI changed how you approach web development—or any other creative work? Let me know in the comments.
This is part of the Make vs Buy series, exploring whether it’s better to build things yourself or pay for ready-made solutions, and the conclusion of Series 1. In the next series, I’ll be looking at common household tasks, and when it’s best to buy a service, or do it yourself.