Retour au blog
Tech & AI Décembre 2025 • 5 min read

🚨 The "AI Will Replace Developers" Prophecy is Failing. Miserably.

Remember the panic? The headlines? The mass layoffs, only to be followed by a massive hiring spree for AI talent? I watched it all with a mix of amusement and concern.

We were bombarded with a powerful narrative: LLMs and AI Agents were coming for our jobs. "VibeCoding" was the new trend, and self-proclaimed experts built armies of AI agents to do their work. It was enough to make anyone in tech sweat! 😅

But then… reality hit. And it hit hard. 💥

The same companies that let developers go were soon rehiring. Why? Because the grand promise hit some very real walls:

🤯 The "Garbage In, Garbage Out" Principle Still Reigns

We quickly learned that code generated, evaluated, and tested solely by LLMs can still be… well, profoundly flawed. Giving them more context wasn't a magic bullet either. After 10 failed attempts to solve a problem, even the most advanced LLMs hit a wall and admit their limits.

🧠 The Illusion of Competence

When tested on rigorous algorithmic benchmarks, LLMs often fail to provide a solution. When they succeed, it's rarely optimal. And when it's optimal, they can't creatively pivot to solve the same problem using a different methodology. This lack of true understanding and adaptability is a fundamental chasm.

⚡ The Energy & Cost Reality

As highlighted in brilliant new research like the thesis of Tristan Coignion – "Empirical Evaluation of the Energy Impact of LLMs for Code Generation and Optimization" – the cost is staggering. An engineer is not just a code-producing machine; we are designers, architects, debuggers, and innovators. LLMs, in their current form, are incredibly energy-intensive tools, not replacements.

So, where does this leave us? 🙏

Initiatives like OnlyDust, which aim to build more powerful benchmarks, are crucial. But let's be clear: LLMs and AI Agents are not ready to replace engineers. The dream of fully autonomous computing remains a monumental challenge. 🏔️

This isn't a time for fear; it's a time for contribution. 🛠️ There is so much work to be done in the field of autonomous systems, and each of us has a stone to add to the edifice.

The future isn't about replacement; it's about collaboration. It's about using these powerful tools to augment our capabilities, not to make us obsolete.

So, to all my fellow engineers: Breathe easy, keep building, and let's shape this future together. 💪✨

#AI #LLM #SoftwareEngineering #FutureOfTech #AutonomousSystems
AZ

Arléon Zemtsop

Research Engineer at Inria Lille • PhD Student • Self-Adaptive Systems

Partager: