1. AI Really Does Make Developers More Productive
  2. AI Is Good at Generating Code, but Generating Code Is Not the Same as Building a Product
  3. Why the Final Stages of Development Can Be the Most Difficult
  4. The Faster We Create Code, the More Important Testing and Control Become
  5. Can You Trust AI-Generated Code?
  6. The Less Someone Understands the Technology, the Harder It Is to Verify AI
  7. AI Can Accelerate Work, but It Does Not Decide What Needs to Be Built
  8. AI Does Not Eliminate Bad Architecture — It Can Scale Its Problems
  9. So, Can AI Replace Developers?
  10. What Does This Mean for Businesses?
  11. The Future Is Not AI vs. Developers
  12. Conclusion

AI can already write code.

It can create interfaces, APIs, SQL queries, tests, documentation, and help developers understand existing code. With modern AI tools, a developer can get in a few minutes what might have taken hours before.

So one question is being asked by many business owners today:

If AI can already program, why do we need developers at all?

In 2025, the DORA team, Google Cloud’s research program, conducted a large-scale study of how artificial intelligence is changing software development. The research was based on responses from nearly 5,000 professionals, more than 100 hours of qualitative interviews, and other data.

But the result was more interesting than a simple “AI will replace programmers” or “AI will never replace programmers.”

The main DORA conclusion:

“AI is an amplifier.”

In other words, AI is an amplifier. It strengthens what already exists within a team and organization: good processes become more efficient, while existing problems can become more visible and scale faster.

What does this mean for a business planning to build a website, web application, or SaaS product?

AI Really Does Make Developers More Productive

Let’s start with the obvious.

DORA found that AI adoption in software development has already become nearly universal.

90% of surveyed professionals reported using AI at work, and more than 80% said that it increased their productivity. At the same time, around 30% reported having little or no trust in AI-generated code.

This is an important point.

The research does not say that AI is useless. Quite the opposite: the data shows a significant positive effect on individual productivity.

Google summarizes the main finding this way:

“AI doesn’t fix a team; it amplifies what’s already there.”

In other words, a developer using AI really can accomplish more.

But that does not mean AI can independently replace a developer.

AI Is Good at Generating Code, but Generating Code Is Not the Same as Building a Product

One of the key changes is that getting the first working version of a software product has become much easier.

AI can quickly help a developer write new code, modify existing code, perform technical analysis, create tests, and complete many other development tasks.

But DORA draws attention to another question: what happens after the first version has already been created?

In a later analysis of the research findings, DORA referred to this as the workflow gap — the gap between rapid prototyping and preparing a product for production.

The researchers write:

“AI can speed up the initial bulk of the work (prototyping).”

However, creating a prototype is only one part of software development.

After that, a project may require integration with existing systems, handling edge cases, validating application behavior, testing, and preparing the product for real-world use.

And this is exactly where the advantage gained from creating the first version very quickly can partially disappear.

DORA describes the effect even more precisely:

“the remainder of the work (production integration) often neutralizes those gains.”

This is a fundamental distinction.

AI can significantly accelerate the creation of an initial working prototype. But production-ready software is much more than an initial working prototype.

Why the Final Stages of Development Can Be the Most Difficult

Let’s take a simple example.

An entrepreneur wants to build an online booking system.

AI can quickly help create:

  • user registration;
  • user accounts;
  • a booking form;
  • a calendar;
  • search;
  • payment integration;
  • email notifications;
  • an admin dashboard.

In a demo, everything may look great.

But a real product has to answer much more complex questions:

— What happens if two people try to book the same property at the same time?

— What happens if the payment goes through but the database record is not created?

— What happens when a request is submitted more than once?

— What happens if an external API is temporarily unavailable?

— How should access permissions work?

— Which data is the source of truth?

— How should the system behave as the load increases?

These are no longer simply “write the code” tasks. They are part of software development.

They involve architecture, business logic, security, reliability, and integration.

That is why DORA emphasizes the importance of maintaining engineering practices even as development speed increases dramatically.

The Google research states directly:

“AI accelerates software development, but that acceleration can expose weaknesses downstream.”

In other words, AI can accelerate the production of changes while simultaneously making existing weaknesses in the system more visible.

The Faster We Create Code, the More Important Testing and Control Become

This is perhaps one of the most important findings of the research for businesses.

It would be easy to assume:

More AI → more code → faster development → faster finished product.

But reality is more complicated.

DORA found that AI adoption in 2025 was already positively associated with software development and delivery throughput and product performance. At the same time, there was a negative relationship with software delivery stability.

Why does this happen?

Because AI allows teams to create changes much faster.

As a result, without appropriate control mechanisms, a team may simply start producing more changes than its processes can safely validate and release.

DORA describes it this way:

“Without robust control systems”

An increase in the volume of changes can lead to instability. The research identifies automated testing, mature version control practices, and fast feedback loops among the necessary mechanisms.

Therefore, AI does not eliminate the need for:

  • testing;
  • code review;
  • version control;
  • continuous integration and continuous delivery (CI/CD);
  • monitoring;
  • automated checks;
  • secure deployment processes.

On the contrary, the more code a team can generate, the more important these mechanisms become.

Can You Trust AI-Generated Code?

This is where the DORA findings become particularly interesting.

AI-generated code can look perfectly normal.

It may successfully pass several simple tests.

It may even solve the task it was given.

But that does not necessarily mean the solution is optimal or secure.

According to DORA, around 30% of professionals reported little or no trust in AI-generated code.

In its subsequent analysis, DORA describes the attitude of professionals toward AI as:

“a ’trust but verify’ mindset.”

This is an excellent description of the modern approach.

A developer does not necessarily have to write all the code manually.

They can let AI generate a significant portion of it.

But they need to understand how to verify the resulting code.

And this is where experience becomes especially important.

The Less Someone Understands the Technology, the Harder It Is to Verify AI

At first glance, AI may make software development more accessible to people who previously lacked deep technical knowledge.

You can give AI a task:

“Create an authentication system for me in Laravel.”

And get hundreds of lines of code.

But what happens if the solution contains an error?

How do you know that an error exists?

How can you determine whether the implementation is secure?

How do you know whether the architecture will support the project as it grows?

How can you verify that AI has not created technical debt that will become a problem a year from now?

This paradox is also discussed in DORA’s subsequent qualitative analysis.

In one interview, a developer described the situation as:

“the blind leading the blind.”

The context is important: AI allows people to work with technologies in which they have insufficient experience, but at the same time, this makes it harder for them to verify whether the resulting solution is correct.

That is why the ability to evaluate AI-generated output is becoming just as important as the ability to generate it.

AI Can Accelerate Work, but It Does Not Decide What Needs to Be Built

There is another fundamental problem.

AI can be very good at completing a task.

But first, someone needs to define the right task.

For example, an entrepreneur says:

“I need a marketplace.”

But a marketplace is not a single feature.

You need to decide:

  • who the sellers are;
  • who the buyers are;
  • how listings are created;
  • what roles exist;
  • how commissions are calculated;
  • how refunds work;
  • what the moderation rules are;
  • what data needs to be stored;
  • what actions are available to each role;
  • what happens in disputed situations.

AI can help implement these rules.

But who defines the rules themselves?

That is a product and business question.

This is why DORA specifically emphasizes user-centricity — keeping the user at the center of development.

Google writes:

“AI becomes most useful when it’s pointed at a clear problem.”

This is a very important conclusion.

The better defined the problem we are trying to solve, the more useful AI becomes.

AI does not replace understanding what needs to be built and why.

AI Does Not Eliminate Bad Architecture — It Can Scale Its Problems

Another important DORA finding concerns not the AI tool itself, but the environment in which it operates.

The research shows that organizations cannot simply buy AI tools and expect the desired results.

They need a strong technical and organizational foundation.

DORA puts it directly:

“Successful AI adoption is a systems problem, not a tools problem.”

In other words, the question is not simply whether you use Cursor, Gemini, Claude, or another AI tool.

The question is how well the software development system itself is organized.

DORA identifies seven capabilities that help organizations gain positive value from AI, including:

  • a clear AI policy;
  • a healthy data ecosystem;
  • connecting AI to the organization’s internal context;
  • small development batches;
  • user-centricity;
  • robust security mechanisms;
  • a high-quality internal platform.

This leads to an interesting paradox:

The more AI changes software development, the more important fundamental engineering practices become.

So, Can AI Replace Developers?

Now we can return to the original question.

The DORA research does not conclude that “AI will replace developers.”

But its findings allow us to make a much more useful conclusion for businesses.

AI can already perform a significant portion of individual tasks that developers previously handled manually.

It can write code.

It can explain code.

It can create tests.

It can help with documentation.

It can accelerate prototyping.

It can help find and fix bugs.

And all of this already has a measurable impact on individual productivity.

But software development is not just code generation.

It also involves requirements, architecture, decision-making, validation, integration, testing, security, and operations.

That is why the more accurate question today is not:

“AI or developer?”

It is:

“Which parts of software development can AI perform independently, and where is engineering oversight still required?”

What Does This Mean for Businesses?

For entrepreneurs, this is probably the most practical part of the entire research.

If you need a simple website or a small prototype, AI can indeed significantly reduce development costs.

If you need to validate a business idea, AI allows you to create the first version much faster and at a lower cost.

But when you are building a system that your business depends on, the situation changes.

For example:

  • SaaS products;
  • marketplaces;
  • e-commerce stores;
  • booking platforms;
  • corporate websites;
  • startup MVPs;
  • customer portals;
  • complex WordPress/WooCommerce projects;
  • applications with payment processing;
  • systems with multiple external APIs;
  • applications with complex roles and access permissions;
  • existing systems that need to be scaled or modernized.

In these projects, the question is no longer whether AI can write code.

It can.

The question is whether that code solves the business problem correctly and how reliably it will work in the real world.

This is where an experienced web developer helps turn a technical implementation into a reliable product that solves specific business needs.

The Future Is Not AI vs. Developers

Perhaps the most reasonable conclusion from the DORA research is that putting AI and developers against each other is not quite the right way to look at it.

AI is already becoming part of the software development process itself.

So the real competition looks more like this:

a developer without AI

versus

a developer who knows how to use AI effectively.

But even here, an AI tool alone is not enough.

As DORA says:

“The value of AI is unlocked not by the tools themselves.”

Value emerges when AI is integrated into a well-organized software development process.

That is why the role of the developer is gradually changing.

Developers spend less time on mechanical code writing and more time on:

  • architecture;
  • technical decision-making;
  • reviewing AI-generated code;
  • integration;
  • testing;
  • security;
  • performance;
  • understanding business logic;
  • turning a prototype into a production-ready product.

For web development, the same principle applies: the value of a professional is increasingly determined not by how much code they write manually, but by their ability to design, validate, and evolve a web product.

Conclusion

AI is genuinely changing web development and software development.

And it is doing so much faster than many people expected.

The first version of a product can now be created relatively quickly.

But this does not mean that software development has become fully automated.

The Google DORA research paints a much more interesting picture:

AI accelerates software development, but at the same time increases the importance of engineering processes, result validation, and the quality of the technical environment.

AI can create code.

But a production system needs to be reliable.

AI can create a prototype.

But a business needs a working product.

AI can suggest a solution.

But someone still needs to determine whether that solution is correct.

That is why the question for businesses in 2026 should no longer be:

“Do I need a developer if AI exists?”

A much more useful question is:

“How can I use AI to get a finished product faster and at a lower cost without sacrificing its quality and reliability?”

And according to the DORA research, this is the direction in which software development is evolving today.

AI makes the first version cheaper, but turning that version into a reliable production-ready product remains an engineering task.

Sources