Codeless: How AI Is Changing Software Development
What happens when you abolish manual coding
Four months ago, one company banned manual coding. Not encouraged. Banned. Most companies are currently somewhere in the middle: a few developers use Cursor or Copilot, and the leadership level calls it "AI-assisted development" and moves on to the next topic.
What is easy to overlook here: the developers who go all in get faster every week. The ones who do not keep up fall behind. This gap grows with every week. At myos, we use AI-assisted development for our own client projects and see this effect firsthand.
Different models, different strengths
AI models bring different strengths. Some read through the entire codebase thoroughly before they write a single line. This takes time, but it considerably increases the likelihood that the right place gets changed.
Other models have a better feel for product and user experience. They are suited for UI, frontend, and everything where it comes down to the interplay of functionality and user experience.
Figuring out which model works best for which task took a few weeks at myos. You suddenly think differently about how to distribute work: who do I give this task to? How much context does it need? How much of the codebase could be affected?
The self-verification problem
In the early phase, models confidently produced code that did not work. Only when running it did the problem show up. At this point, many teams give up and retreat to optional adoption.
The solution lies in infrastructure that forces models to verify their own work before a task is considered done. At myos, we have developed our own verification pipelines for this. It requires upfront effort, but it is the step that turns unreliable AI output into dependable results.
The economics of experimentation
The clearest proof: a system with around 90,000 lines of code, built in about two weeks. A single developer would traditionally have needed almost a year for this.
But the sheer volume is not the point. What has changed is the economics of experimentation. Cloning and customizing a client app: from weeks to hours. Testing an implementation idea: an afternoon instead of three meetings. When the cost of experimentation drops this much, better results emerge, because more variants get tested.
At myos, we use this directly for our clients: prototypes in days instead of months, validating ideas before large budgets are released, improving solutions iteratively.
What matters now
The developers who adapted the fastest were not the ones with the deepest language expertise. They were the ones who already knew how to formulate tasks clearly, provide enough context, and evaluate results without rewriting everything themselves.
The central competency in engineering is shifting. It is less about the how, because that is increasingly handled by the models, and more about the what and why: product understanding, architectural intuition, judgment. Those are the skills that AI does not replace and that now determine the quality of the output.
