How AI is Changing Development: From Skepticism to Real Results

A year ago, if someone had asked me: "Can you build a proper application with AI?"  I would have categorically answered "no." Today, my answer has changed to "yes," but with important caveats. Over this past year, we've gone from experiments to real AI implementation in our development process, and the results have been surprising.

Where AI Actually Works

MVPs and greenfield projects — this is where AI tools show their best performance. When there's no technical debt, outdated dependencies, or convoluted architecture, AI can generate clean, modern code following current best practices. You can get a working prototype in a few hours that would have taken a week before.

According to GitHub's 2024 data (and things have only improved since then), developers using Copilot complete tasks 55% faster. Anthropic reported in 2024 that Claude Code reduces time spent writing boilerplate code by 60-70%. These aren't just numbers — this is real time saved on routine work.

Legacy projects — that's a completely different story. AI doesn't work miracles here yet. Old code written by several generations of developers, with its own patterns and peculiarities, often stumps AI. But even here there are applications:

• Quick orientation in unfamiliar code
• Targeted debugging of specific functions
• Writing tests for existing logic

The Experiment That Changed Our Approach

A few months ago, we decided to try something unusual: we asked one of our business analysts not just to describe requirements for new website pages, but to code them herself using Claude and other AI tools.

It sounded ambitious. Someone without development experience was supposed to create production-ready code.

The first steps were difficult:

• Setting up the IDE required developer help
• Environment setup proved too much even for Claude
• Three sessions with a developer went into initial configuration

But then something interesting happened. After overcoming the entry barrier, our BA started creating quality landing pages. Not prototypes, not drafts — full-fledged pages that passed code review and made it to production.

The process looked like this:

1. BA writes code with AI assistance
2. Commit goes to team lead
3. Code review by all standards
4. Merge to main branch

No compromises on quality. No "good enough for a BA." Standard pipeline, standard requirements.

Impact on Client Estimates

This shift has tangibly affected the estimates we provide to our clients. For greenfield projects, we're genuinely saving about 50% of the time we would have spent without AI. This isn't marketing speak — these are real hours that we can now either reinvest in quality or pass savings to clients.

However, it's crucial to emphasize: this efficiency gain applies almost exclusively to projects started from scratch. Legacy codebases don't see the same dramatic improvements.

What This Means for the Industry

This experiment revealed several important insights:

AI doesn't replace developers. We still need engineers for:

• Architectural decisions
• Complex business logic
• Code review
• Environment setup
• Mentoring and knowledge transfer

AI democratizes development. People without technical backgrounds can create real code. According to Stack Overflow Developer Survey 2024 (these trends have only accelerated since), 76% of developers were already using AI tools at work, and 70% believed AI makes programming more accessible to newcomers.

The entry barrier has shifted. Complexity moved from "how to write code" to "how to properly formulate the task" and "how to verify the result." This fundamentally changes developers' roles — from executors to architects and reviewers.

Real Savings

According to McKinsey's research, companies that implemented AI in development report:

• 30-40% faster delivery for new features
• 50-60% reduction in time spent on routine tasks
• 15-25% reduction in bugs thanks to AI-assisted testing

In our case, we no longer need to wait for a developer to become available for simple landing pages. BAs can independently implement their ideas, while developers focus on complex challenges. 

Conclusions

AI in development isn't hype and it's not replacing people. It's a tool that:

• Works excellently for quick-starting new projects
• Helps optimize routine work in large projects
• Allows non-technical specialists to make technical contributions
• Requires expertise for proper application

A year ago, I didn't believe in AI development. Today, I watch as our business analyst commit code to production. This doesn't replace developers — it expands the capabilities of the entire team.

The key is understanding: AI is an amplifier, not a replacement. And like any tool, it's only effective in the right hands and for the right tasks.

The 50% time savings we're seeing on greenfield projects isn't just changing how we work — it's changing how we price projects and what we can promise clients. But this efficiency gain comes with a caveat: it only applies when starting fresh. For legacy systems, the old rules still apply.