How AI Shortens Development Time (and Where It Doesn’t Help At All)

And yet, it doesn’t work that great in all scenarios - there are still pitfalls and risks we need to be aware of. It’s risky to assume AI will do the work for you. As many of us have noticed, sometimes it takes longer to get AI to do exactly what we need than to do it ourselves.

Once it solves very complex issues, and the other time it fails on something potentially simple.

Some studies actually show that AI can make us slower, not faster. It was a study conducted in 2025 on experienced developers measuring productivity with AI. Before starting tasks, developers forecasted that AI would reduce completion time by 24%. After completing the study, developers estimated that AI reduced completion time by 20%. Surprisingly, the study found that allowing AI actually increases completion time by 19%.

On the other hand, LLMs are developing rapidly, and we can see a change in the generated code in just a year. Are studies from a year or two ago still relevant? It’s hard to say. One sure thing is that AI is changing things so rapidly that we all need to try hard not to stay behind.

So, will AI make development faster in 2026? Yes. Probably. Usually. Sure, unless…

How It Can Make Us Slower

Well, sometimes you spend a significant amount of time writing a detailed and well-prepared prompt. The model thinks, thinks, thinks… and then it comes up with something that doesn’t work at all. Then you spend even more time going back and forth with the model. At some point, it works 90% of the time, but there are still things AI can’t get right. And you think: “I should have written it myself.” Sounds familiar?

The bigger the system, the harder it is. Different models have different context windows. Context is the total relevant background information - including prompt, conversation history, and data - that the model processes to understand and generate accurate responses.

In practice, this often means:

• missing or incomplete context
• oversimplified assumptions about the system
• edge cases that don’t fit the model’s reasoning

Also, if we rely on AI too much to write our code and accept everything without even reading it, then we stop understanding our system. We don’t really learn. So if AI struggles with a working solution (due to complexity, lack of context, or edge cases) we’re at a loss. A developer who relies heavily on AI and doesn’t know the ins and outs of their system will lose more time fixing problems than one who wrote the code themselves.

How It Can Make Us Faster

But obviously, it does speed up many things. Apart from implementing new features in mere minutes, debugging and translating designs to HTML code on the fly, it does help with tasks that are not developing per se, but are still a core part of a developer’s work.

AI works especially well for:

• setting up basic project structure
• internal tools and back-office systems
• quick early prototypes and first versions of apps

Research

AI helps quickly research possible solutions. It always requires validation, but it can definitely give good ideas or push you in a new direction. You still need to verify it by yourself, but this is most helpful when dipping into completely new things - it’s hard to even start research if you don’t know what to look for, and AI will likely do it for you.

Prototyping and Requirements Gathering

We saw this work in real life with our latest project. We were gathering requirements for a small custom management app, tailored to our client’s business needs. The client’s industry was completely new to our team, so we couldn’t predict all the real-life problems they were facing. It was a back-office system, and the client had no requirements for the app’s look, so there was no need for design or Figma prototypes.

Instead of preparing the design, we started building the app early. We created a simple proof of concept in several hours using React Admin and Firebase, and we let Claude Opus do most of the work. Then we showed it to the client. We immediately got valuable feedback. Some requirements weren’t obvious at first, but the client could test the working prototype, help us identify pitfalls and missing features. The client showed us precisely what they needed.

It’s not a new idea - I’ve worked on startup projects with demo-driven development: create a prototype, get stakeholder feedback, rewrite, repeat. But AI makes the initial prototype much faster, opening it up to more clients and industries. And this prototype doesn’t need to stay one; it can become the base for the real project.

Estimations

Recently, I prepared an estimation for a new project myself, then asked AI to verify it. In this project, we couldn’t rely on any ready solution. Business logic was too complex and there were many functionalities that required a custom approach. I got valuable feedback and a list of edge cases I didn’t think of. AI probably didn’t make the development itself faster, but it let us know sooner what could be more time-consuming. In other words, it helped us to prepare better.

Estimates are not easy, especially if you are estimating not a single task, but the whole project. Underestimating will end up in stress and lost money, for you and for the client. Overestimating may cost you the lead.

This is probably my least favorite part of the developer’s job, and I’m sure I’m not the only one. If I can get any help, I will gladly take it. Of course, you need to be cautious with the number

Conclusion

One thing is sure: development isn’t what it was a few years ago, but it still requires human touch. There’s this theory of model collapse - a degenerative process where models trained on AI-generated data lose accuracy, diversity, and quality over time. It means that models need human input to develop and that there is a big value in writing real code, filing good bug reports, and documenting tricky behavior.

AI speeds up many things, but we need to keep thinking, questioning its output, and creating things ourselves.

AI is a powerful accelerator but only for developers who still understand what they’re building.