Software has already eaten the world, and systems are becoming increasingly complex.

While AI-assisted tools like Copilot and Cursor are excellent in the hands of a skilled developer, they won’t, by themselves, solve this growing challenge. Software, and the documentation that comes with it, starts to decay the moment it’s created. It requires continuous care and maintenance, yet there’s rarely as much time for this as the software truly demands. Inevitably, this leads to the accumulation of technical debt.

Right now, we need to shift our mindset: large language models should be used to clean up the mess. This is the kind of heavy, neglected work that tends to get overlooked from one project to the next. When used correctly, today’s tools can already help pay down technical debt, keep documentation up to date, and even engage in meaningful conversations about the systems being developed. All of this reduces the cognitive load on developers and the wider organization, allowing them to focus more on work that actually drives new value. In my view, this is poised to be the next major breakthrough in software development.

So, how do we get there? Even with modest investments, organizations can elevate their AI capabilities to a level where they start paying for themselves. This means leveraging existing tools specifically to reduce complexity. You’ll find concrete examples of this in the related articles listed to the right. And if you’d like to continue the conversation, don’t hesitate to book a time. I’d love to chat!