AI has made it dramatically easier to turn an idea into working software. The harder question is what happens when that weekend project becomes a system people depend on.
Not that long ago, building a software product required a fairly predictable journey.
You needed an idea. Then came requirements, wireframes, developers, infrastructure, testing and deployment. Depending on the project, it could take weeks or months before anyone outside the development team could actually use the product.
That is changing.
Today, someone with an idea can sit down with an AI coding assistant and start building. They describe what they want. The AI generates code. They test it. They make changes by describing what isn't working.
A few hours later, they have something real. A few days later, they might have an entire application.
Welcome to the era of vibe coding.
The barrier between an idea and working software has become dramatically smaller
This is probably the most interesting part of the AI revolution for me.
We're no longer just using AI to answer questions or generate pieces of code. We're increasingly using it to help turn ideas directly into working software.
- Someone who may never have considered themselves a developer can now build a prototype.
- A founder can validate an idea before investing heavily in a development team.
- A developer can build in hours what may previously have taken days.
The speed is remarkable.
But there's an interesting point where the nature of the project changes. At some point, it stops being an experiment. Someone wants to use it. Then another person signs up. The data starts becoming important. People expect it to be available.
Suddenly, that project you built over a weekend has become a system.
This is where things start getting interesting
Getting an application to work is one challenge. Running it reliably is another.
- Where should it run?
- How is it deployed?
- What happens when you make a change?
- Where does the data live?
- What happens if something fails at 2 AM — and how do you know that it has failed?
- How do you back it up?
- How do you secure it?
- And perhaps most importantly: how much should all of this actually cost?
These aren't necessarily the questions you think about when you're excitedly building version one of an idea. And perhaps you shouldn't. The ability to move quickly is one of the biggest advantages of modern development.
Not every weekend project needs enterprise infrastructure. Not every new application needs Kubernetes. Not every startup needs a complicated architecture capable of supporting millions of users on day one. Sometimes, the best infrastructure is simply the infrastructure you don't need yet.
The challenge is knowing when things need to change
An application serving ten users has very different requirements from one serving ten thousand. A prototype has different requirements from a business processing customer data. A small application can often run perfectly well on surprisingly simple infrastructure.
But as usage grows, the questions start changing.
- Reliability starts mattering more.
- Security starts mattering more.
- Backups become less of a "we should probably do that" task and more of a necessity.
- Deployments need to become repeatable.
- Monitoring becomes important.
- Costs need to be understood.
This is where I think the next interesting phase of vibe coding will emerge. AI is making it easier than ever to build software. The next challenge will be making it easier to run software.
The future might not be about writing less code
It might be about spending less time worrying about everything around the code.
Imagine being able to take an idea, build an application, deploy it safely, monitor it, secure it and scale it without needing to become an expert in infrastructure along the way. We're not entirely there yet. But we're moving in that direction.
AI has dramatically changed the beginning of the software development journey. I think we're about to see it change the rest of the journey too. And that's going to be fascinating to watch.
The ability to build software is becoming more accessible. The responsibility of running that software isn't disappearing — it is simply becoming the next problem to solve.
What do you think? Have you built something with AI that went from a quick experiment to something people actually started using?