When I see projects like gpt-pilot, I can’t help but think they will accelerate the pace at which we produce systems today.

An intelligent person with minor to no knowledge of development can produce some exciting systems with such tools; one may be able to sell this thing to a few people and make it an essential part of many other businesses.

Yet the systems produced by these AIs are exposed to the same problems as any software, BUGS. I predict that in the future, there will be many businesses selling systems created by AI, and the founders of those businesses will need experienced developers to fix these complex issues.

What issues this system may have?

This would be a great topic for a master’s degree student. I will document out of hand a few I think will happen in this blog, and I can come back to this 10 years from now and see how right or wrong I was.

1. Performance issues

We all know for a fact that 99% of the code out there is not performant nor made for performance; in many cases, it is actually code made in a rush for a POC or to learn something. It is reasonable to assume that the code AIs are learning from code that is not made for performance. Most systems can support a few hundred users without issue.

When some of those apps came to success, I predict most will have performance issues. The context needed to reason what to do in this kinda situation, what to improve, and what does not need to improve may be too big and will be the time for experienced engineers to fix.

2. Edge cases bugs

Some bugs occur only once in every million transactions, involving edge cases and race conditions that require extensive context—billions of tokens—to even begin to understand the issue. Sometimes, it is more an art than a sense of what to check and how to fix. I predict AI won’t be able to find those or fix them, and that will be the job of very experienced developers to come to dig logs and metrics for weeks to see the two lines of fixes needed to fix this kind of bug.

3. Time to marketing for new features will reduce over time

The apps fully developed by AI will require a greater and greater context to extend and maintain, which will make it harder and harder to add new features. At some point, it will stop progressing and will require experienced developers to come to the game, find the reasons why it is getting harder, and provide simple contexts for the AI to continue to progress.

Conclusion

AI will open the door for much more software to be produced and may enable smart people to bring their ideas to life and test if the product is viable. This will create opportunities for experienced software developers to assist these systems and businesses in progressing when issues arise. My prediction is that our job as developers will be to create the infrastructure that will support this app (Like NunDB) and fix the complex issues that AI will not be able to repair or find; in a few years, we will know if I am right.