Clément Hervé

The real danger of AI with software

When people think about the danger of AI, it might be about "it's going to replace us", or "it's going to pwn all the software in the world". Even if there might be a slight ounce of truth in all of these claims, they are not really relevant. The worst thing that could happen is if we lose our craft and are unable to think for ourselves. If this day comes, we are thoroughly f*cked. Because if this day come, it means we have surrendered to Anthropic, OpenAI and the other tech giants and they now can control the software.

People often tell we are "only going up the abstraction ladder" like we did with compilers. We're not. Everyone claiming that today is a moron. Compilers are deterministic and most often open-source and they don't require a $100K hardware rig to operate. If you give up your skills to write software, you are only a serf to the all powerful Silicon Valley and they own you. They can raise their prices, lower their quotas, or do whatever they want with your code.

If you want to protect your independence, work with humans. You can still use an AI to speed up your tasks, but it still must be a 1-1 mapping and humans must know the code, know how to debug it and how it works.

So how do we do that? As a passionate software engineer, I've worked more and more with AI to help me in my workflows. I divide tasks in three buckets:

I also always start by drafting up a plan with the AI to both save tokens, take the time to match my mental model with the model the AI has built and be confident of the implementation before hitting enter.

When there are tough or new topics, I'll ask the AI to write in my Obsidian Vault a note, tagged with the date and other relevant informations so I can come back to it from time to time and make sure my memory is intact.

On my free time, I do Leetcode to avoid losing my skills.

And this is how I use AI to speed up my tasks all the while maintaining my skills.