How AI is changing software development right now

Right now, four changes are happening at the same time. Each of them affects how software should be built, how developer productivity is evaluated, and how to achieve results that actually last.

1) Code generation: AI as a co-worker, not just a tool

GitHub Copilot, Cursor, and Claude have moved beyond simple autocomplete. They are already taking part in genuine problem-solving. In many teams, a significant share of the code that reaches production is now created with AI assistance or generated by AI. We are constantly being asked about AI as well.

That is not the most interesting part, though. More important is what all of this requires from the person sitting in front of the machine.

In practice, the change looks like this:

• Give AI the full context: the goal, the constraints, and the existing architecture
• Review its output with the same care you would apply to a junior developer’s pull request
• Combine its speed with your own understanding of what the system actually needs to do

Speed is useful only when the direction is right. Defining that direction still requires professional judgement and deep expertise.

2) Thinking: the new bottleneck in software development

When AI takes care of the mechanical work, the bottleneck shifts from writing code to evaluating it. Someone still has to decide whether the code is the right kind of code, built in the right way, and for the right reasons.

Poor judgement at AI speed produces poor results at the same speed. The cost of a bad architectural decision is now higher than before, because new layers can be built on top of it very quickly.

What is worth investing in now:

• Code review – the ability to spot problems before they reach production
• Architectural thinking – understanding what should be built and what should be left out
• Testing mindset – the ability to define what “works” means before a single line of code is written

These ways of thinking and skills increase in value over time. Simply refining prompts does not do that. Nor does blindly trusting AI-generated output without developing your own expertise alongside it.

3) Domain expertise: a capability AI cannot replace

As implementation speeds up, value moves upstream: to understanding the problem deeply enough to guide the tools in the right direction. AI does not create domain context out of thin air. It can summarise what is already known, but it still needs a human alongside it who truly understands the field and can, for example, think through usability.

This is good news for people with deep specialist knowledge. A developer who understands industrial processes, the constraints of logistics, or financial regulation is now in a stronger position than a colleague focused only on syntax.

The combination that makes the difference:

“I understand this industry and its real-world constraints”
AND
“I know how to guide AI to build solutions that truly fit it”

That combination is still rare. And increasingly, it is exactly what projects need.

4) Team size: smaller teams, greater output

One developer or a two-person team can now deliver what previously required ten developers. AI acts as a force multiplier across the entire stack: it builds the foundation of the software, writes tests, produces documentation, and handles repetitive refactoring.

As a result, the economics of software are changing as well. The barrier to building is getting lower. That is good news for individuals and small companies, but at the same time a real strategic question for larger organisations that have been built around headcount.

So what should be done to make the future as strong as possible? The right answer depends on your situation:

• Individual developers: use this phase to broaden your skills across the full stack, not just to deepen one layer
• Team leads: reassess the right team size and the right mix of skills for each task
• Organisations: invest in people who know how to work with AI, not merely around it

The developers who succeed five years from now will probably not be the ones who resisted AI. Nor will they be the ones who outsourced their thinking to it completely. The ones who succeed will be those who learn to use AI effectively while continuously developing their own expertise.

My question to you: what is one concrete thing you are doing right now to keep up with this change.