It’s a strange and exciting time to be a software engineer. And a software engineering client.
AI is changing the pace of software development in a very real way. Our team can move faster and do more with less. Prototypes take minutes or hours instead of weeks or months. We can test ideas earlier and iterate more within the same budgets that once only got us to a functional MVP.
These are wins, especially for a team that loves technology and wants to do as much good as possible within the time and budgets available to us. But the speed and accessibility of AI are also shifting client expectations.
With AI automation, organizations assume digital experiences should be faster, smarter, and more personalized—not to mention significantly less expensive to build. And of course that makes sense. If AI can accelerate software development processes, shouldn’t software become easier and cheaper to build?
In theory, yes. But this assumes software developers are solely focused on generating code, which has never been the case.
AI is incredibly good at accelerating execution. It can help automate repetitive tasks, speed up implementation, generate documentation, support testing, and reduce a lot of the friction that slows work down. Used well, it allows us more room to experiment, pivot, and improve our work.
What AI does not do well is understand organizational context, navigate competing priorities, or make thoughtful decisions about what truly matters to users. It cannot balance business goals against technical constraints, operational realities, long-term maintainability, and human behavior. Those decisions remain deeply human—and they are often the difference between software that functions and software that creates lasting value. It’s a theme that recently showed up in Pope Leo XIV’s encyclical on AI. He frames the technology as a valuable tool that demands human vigilance, not a replacement for human judgment.
We recently saw this play out while refactoring a complex, multi-part form central to one of our clients’ operations. An engineer on the project used Claude to quickly uncover historical code comments that documented long-standing performance issues, and then refactored the form to improve how data was saved and edited. Work that could have taken much longer to investigate and understand was accelerated significantly by the use of AI.
But another engineer on our team had prior insight into how users interact with the form and identified opportunities to further improve the experience. Instead of requiring folks to repeatedly stop and click a save button, the team explored auto-saving functionality that lets users work on the form without interruption. The final solution benefited from both approaches, and neither would have been as effective on its own.
This isn’t just our experience. A recent JAMA Network Open study tested 21 leading AI models on clinical reasoning and found they scored well when handed a complete picture — but stumbled badly when the task required holding multiple possibilities open and deciding what to rule out. The models tended to lock onto the obvious answer. Naming the likely diagnosis is the easy part. Knowing what else to consider is the hard part — and the human one.
Human expertise is and will remain integral to every software development engagement. Clients don’t hire Software for Good to crank out code as quickly as possible, but to ensure a high-quality technology solution where strategy, discernment, systems thinking, empathy, and judgment are core to the process. They trust us to know where automation adds value, where it introduces risk, and how to make those calls responsibly and transparently.
As implementation gets easier, that judgment only gets more valuable — not less. AI may change how we build software. But human expertise still determines what software is worth building in the first place.