Just because you can, doesn’t mean you should.
Before you put another thing
on the internet,
give it a reason to exist.
An app. A plugin. A wrapper around a wrapper. AI can help you make it. That still leaves a few good questions: who needs it, what does it add, and who looks after it?
Making something to learn, or just for fun, is a perfectly good reason. Publishing it for other people is a bigger promise.
A few things worth thinking aboutThe easy part
got easier.
Generating code can feel like progress before you’ve worked out what needs fixing. A polished demo doesn’t answer the question: “Would this help anyone?”
Before starting another project, try the existing tools. Talk to the person with the problem. A useful contribution to something good may do more than another version of it.
“It works”
isn’t enough.
Getting it to run is the visible part. How it works, how well it works, and how it fails are what people live with.
A login can work while private records remain accessible to the wrong person. OWASP ranks broken access control first in its 2025 Top 10. A working interface can hide a serious security failure.[3]
In a small 2023 study, people using an older AI coding assistant produced insecure solutions more often on four of five security tasks, while being more likely to believe their code was secure.[2]
The person taking the risk
might not be the person who clicked generate.
You can skip the hire.
Who does the job?
A good developer helps shape the idea, questions the brief, finds a better approach, and investigates why things break. If you build alone, those decisions land on you.
They may use AI too. You’re paying for judgment, creativity, and problem-solving as well as the code. Here’s what to discuss when agreeing the work.[4]
| The part beyond the code | You + an AI tool | You + an experienced developer |
|---|---|---|
| The idea | You spot the real problem and judge the suggestions. | Someone to challenge the brief and explore alternatives with you. |
| The rough edges | You find the bugs, missing cases, and misleading answers. | Someone to investigate causes, test changes, and explain trade-offs. |
| The responsibility | You arrange security review, updates, and support. | Named deliverables and a support plan, if included in your agreement. |
Human work needs vetting too. Ask for relevant experience, clear deliverables, and an explicit testing and support scope.
The subscription
isn’t the whole price.
An editable example in USD. Your time starts at $25/hour, rounded up from the U.S. median.[11] The other inputs are illustrative.
$60 in tools
+ 50 hours of your work × $25/hour
$1,500 developer quote
+ 5 hours of your briefing & review × $25/hour
Including your time, hiring costs $315 more and frees up 45 hours of your time.
Change the assumptions
Totals include cash spending and the value of your time. When hiring, your time covers briefing and reviewing the work. Hosting, ongoing maintenance, and incident costs are excluded on both sides. Actual quotes and skill vary.
And then there’s the month after launch.
If your product calls paid AI models, more usage, a pricier model, or changed pricing can raise the bill. Retired models can mean migration work. Budget for that dependency. A regular site merely written with AI doesn’t need a model subscription to keep running.[7]
Someone did
the hard part.
Before prompting a copy of an existing app or plugin, ask what you’re improving. Consider buying it, supporting its maintainer, or contributing something useful. Their decisions and care have value beyond the interface you can reproduce.
The economic stakes are real: one study found about a 21% relative drop in writing and coding job posts on a freelance platform after ChatGPT launched. That measures demand in one setting, not the effect of every clone.[5]
A copy of a copy needs an original.
Researchers have shown how repeatedly training models on earlier models’ output can lose rare patterns and degrade results: model collapse. It is a training-data risk, not an inevitable result of using AI. Keeping original data matters.[6]
Cheap to click.
Someone pays.
There’s a physical world behind the output. Virginia’s legislative researchers warned that rapid data-center growth could raise other customers’ electricity costs and leave them paying for infrastructure if expected demand falls short.[8]
These are wider infrastructure risks. Data centers serve more than AI; a single project doesn’t explain an electricity bill.
Big Tech data centers are driving up power bills at America’s Rust Belt factories
(opens in a new tab)Ohio suspends data center tax break as tech firms face pressure to pay the cost to power AI
(opens in a new tab)As electricity costs rise, everyone wants data centers to pick up their tab. But how?
(opens in a new tab)Rising AI demand boosts case for new US trunk power lines
(opens in a new tab)Data Centers in Virginia
(opens in a new tab)Make something
worth making.
For someone. For a real problem.
For the joy of learning how.
“Because I can” isn’t enough.