Code smarter, not harder: Using GenAI correctly in software engineering

Good developers cannot be replaced, but they can be supported by AI – with co-pilots that take over routine work and improve quality.

The discussion about GenAI in software development often fluctuates between two extremes: on the one hand, there are sceptics who are dissatisfied with the results achieved so far; on the other, there are visionaries who already believe in fully automated software. The reality lies somewhere in between – and demands excellent developers more than ever.

GenAI in software development

Clean code, automated testing, continuous integration and DevOps have long been standard in modern software projects. Nevertheless, developers spend a lot of time on repetitive tasks – from writing boilerplate code and maintaining documentation to processing simple tickets. This is exactly where generative AI comes in: not as an autopilot that develops independently, but as a co-pilot that takes over recurring tasks – without relieving humans of their responsibility.

Common IDE plugins such as Github Copilot, Amazon CodeWhisperer and JetBrains AI Assistant are designed to enable productivity gains without relieving developers of their responsibility. These code assistants create space for what really matters: architectural questions, design decisions and creative problem solving – the core competencies of experienced developers. AI takes care of the tedious task of writing code.

To use AI effectively in the development process, you need to be able to program yourself – with everything that goes with it: making mistakes, debugging and really understanding the code. Only those who have gone through this learning process themselves can assess when AI really helps – and when it doesn't.

Ultimately, even before the AI era, hardly anyone wanted to be a ‘code monkey’ who just writes uncreative code and maybe does some debugging. What becomes interesting after the first few years of training are high-level topics, the big picture, and challenging decisions that reconcile issues such as environment, budget, skills, requirements, security, performance, and usability with the code. Or designing highly sophisticated new algorithms and methodologies. Writing code and understanding architectures/patterns is a prerequisite for using AI efficiently and safely in the development process.

Quality over quantity

A common mistake when using GenAI is to get excited about producing more code. But more code does not mean better software. The best teams know that quality can be crucial for business-critical software. Experienced developers do not celebrate ‘more’ code, but ‘better’ code. A successful commit can mean fewer lines, the same function – and higher quality.

Proper application also means critically reviewing outputs. No GenAI tool can replace code reviews, pair programming or well-defined quality gates. It is a tool in a process, not a replacement for the process.

However, it is also important to engage intensively with AI assistants, understand prompting and how an LLM works, and be fully familiar with integration into your own IDE.

This is a learning curve that takes time, and efficiency gains are not immediate. Developers need to develop a kind of intuition for when AI is no longer helpful and it is better to continue working in the traditional way. With this knowledge, they can focus more clearly on improving the solution and developing reliable, secure software.

What matters

If you want to use GenAI successfully in software engineering, you should adhere to a few principles:

• Focus on developer experience: Select tools that can be seamlessly integrated into existing workflows and automate repetitive tasks. This requires experimentation to find out what works and what doesn't in your own environment.
• Enablement: Train developers on how to review, question and improve GenAI suggestions. Organise experience exchanges to share knowledge and learn from the mistakes and successes of others.
• Expertise as a success factor: Good results can only be achieved through a deep understanding of architecture, debugging and software design.
• Ensure quality: AI-generated code must also be tested, reviewed and documented – perhaps even more thoroughly. The responsibility remains with humans.

Once you have practised this, GenAI can reduce the workload for developers. This leaves more time to build resilient software, implement additional features and increase stability.

Thomas Dohmke (CEO of GitHub, the manufacturer of Copilot) recently commented that the ‘smartest’ companies are now hiring more developers, not fewer. However, it is important to establish an AI mindset among employees in order to use AI optimally and to recognise use cases where AI helps rather than hinders.

Conclusion

Good developers are not being replaced – they are being relieved of some of their workload. Good, reliable software requires creative problem solving and technical expertise. GenAI is not a substitute for human expertise – but a catalyst that makes technical skills and creative problem solving even more effective and simplifies the process of leveraging these strengths by reducing the burden of routine work and opening up new quality potential.

Or to put it another way:Code smarter, not harder. With GenAI as your co-pilot.

How your team can use GenAI productively – we help with enablement, tool selection and governance.