If LLMs start communicating with each other autonomously and develop their own internal language, the quality of human-readable code may become less important. In fact, humans might no longer be able to judge whether the code itself is good or bad.
I think future programming may look something like this: we tell a conversational interface what we want to build or what problems we have, and it acts as a front desk and translator, passing our intentions to a design-focused LLM that handles the implementation.
In a way, it would come full circle and start to look a lot like today's no-code and low-code tools.
> On the other hand, there's no reason to believe that, in this hypothesized future, LLMs would stick to the languages that exist today. With the exception of machine code, all programming languages exist to make programming easier for humans. If no humans look at code, it's conceivable that new languages will emerge; languages optimized for LLMs to consume, generate, and manipulate.
Yup, been thinking this for a while now.
Also, separately, that while we are in the "centaur" domain, it's important to focus on tools to make it easier to review code and code changes.
> Future LLMs write code in LLaMe. Do they accrue technical debt?
> Honestly, I don't know.
I believe technical debt to be inevitable. At least for the foreseeable future, as the foreseeable future has an ongoing drift from what is made to solve a need in one moment and the changing needs that come later.
And we should also look at the security perspective, Devs with AI produce 10x more "security findings" (https://www.theregister.com/2025/09/05/ai_code_assistants_se...)
Another report says that AI code has 3 times more vulns then human ones with 45% being related with the OWASP top 10 (https://www.veracode.com/blog/genai-code-security-report/)
your second study is a year old, a bit much in an industry moving this fast.
If LLMs start communicating with each other autonomously and develop their own internal language, the quality of human-readable code may become less important. In fact, humans might no longer be able to judge whether the code itself is good or bad.
I think future programming may look something like this: we tell a conversational interface what we want to build or what problems we have, and it acts as a front desk and translator, passing our intentions to a design-focused LLM that handles the implementation.
In a way, it would come full circle and start to look a lot like today's no-code and low-code tools.
> On the other hand, there's no reason to believe that, in this hypothesized future, LLMs would stick to the languages that exist today. With the exception of machine code, all programming languages exist to make programming easier for humans. If no humans look at code, it's conceivable that new languages will emerge; languages optimized for LLMs to consume, generate, and manipulate.
Yup, been thinking this for a while now.
Also, separately, that while we are in the "centaur" domain, it's important to focus on tools to make it easier to review code and code changes.
> Future LLMs write code in LLaMe. Do they accrue technical debt?
> Honestly, I don't know.
I believe technical debt to be inevitable. At least for the foreseeable future, as the foreseeable future has an ongoing drift from what is made to solve a need in one moment and the changing needs that come later.