From builders to confidence builders

Software developers build things. Building is about creating something by writing text, something that ultimately people can interact with. in the era of Ai, machines are also capable to build sometimes they can build fast, and sometimes much faster than humans. This changes things, in a way. Already in 2025, it has become cheaper to prototype things by generating code via LLMs. As those systems become better, they will not be perfect, but it will be cheaper and cheaper to generate code.

Don’t get me wrong. There are tons of risks from a secure perspective, maintainance perspective, and so on. Those risks obviously matter and they will shape a lot of the conversation around code generation with those tools. That said, building prototypes and generating code has become cheaper.

I don’t see a world in which code generation will not be part of it, in a shape or another. In that context, we have to ask ourselves, how has the job of a software developer changed? There is probably a lot of typing, a lot of coding that will disappear in the way we think about it today, but that was not the most important part anyway.

I think the most important part has always been before coding and after coding. Understanding what to build is something a machine cannot really do because it definitely lacks the entire understanding of the business, of the word in which we live in. After the coding part, there is the part about shipping changes, or just building confidence. I think in the code generation era that part becomes even more important and longer because we have less confidence in the code that has been generated.

People that work in DX and infrastructure will not have less to do, they will have more to do, because we are now going to be dealing with code that could be not working, not only according to the spec, not only from a security point of view, but just basically from any point of view.

In this context, the role of the human is to be the ultimate confidence builder, confidence that a piece of code works, confidence that you can ship it, and then that it can be removed quickly from the system if something is wrong.

In the context, the job of platform builders and the DX teams is even more important. Security is also something very central. Countless of times we have seen LLMs tools pick up libraries that don’t exist, that aren’t safe, and using old version of software. Those things will improve too, but we cannot count for those problems to be eliminated because they are intrinsic of how those systems work. The current solution in the industry is that you have one agent reviewing the work of another agent that is doing security on the work of another agent, but that is only going to get you so far.

The ultimate verification and the ultimate confidence builder will be the human, and the human is the one that at the end of the day will also be called accountable and that will be the most important part of the picture, just with less typing involved.