5 Software Turning Points Predicted by Hugging Face's Co-founder

Thomas Wolf published a pretty provocative piece. His argument: in an era where AI writes code, the very structure of software gets flipped upside down. After reading it, I couldn’t stop thinking about it for a while. Some parts I agreed with, others felt like a stretch, so I decided to break it all down.

The Era of Piling On Libraries Is Coming to an End

Until now, it was second nature to pull in packages built by others. Writing everything yourself took too long. But when you can hand things off to an AI agent that builds from scratch, going custom becomes realistic. Fewer external packages means fewer security holes, smaller app sizes, and faster execution.

Working with Claude Code lately, I’ve noticed my npm dependency depth shrinking dramatically. It feels like we’re heading toward a “no dependency” era.

• The chain-reaction vulnerability structure (where one compromised package endangers thousands of projects) starts to disappear
• Smaller bundles lead to faster boot times and response speeds

The Days of “Don’t Touch Legacy Code” Are Over

You’ve probably heard of the Lindy effect: the idea that technology that has survived a long time has good reason to keep surviving. The reluctance to touch legacy code follows similar logic: you never know what might break.

But if AI can read tens of thousands of lines of code and rewrite it in a different language, that logic weakens. Wolf honestly acknowledges one limitation, though: AI still misses unexpected bugs and edge cases. That’s why formal verification (mathematically proving that code behaves as intended) becomes not a nice-to-have but a prerequisite.

• The time and cost to rewrite legacy code has dropped to less than 1/10 of what it used to be
• Deploying AI-written code to production without formal verification is still a gamble

Languages That Are Hard for Humans Are Easy for AI

The popularity of programming languages has always been driven more by psychology than pure technical merit. Is it easy to learn? Is the community welcoming? Will it help me get a job? LLMs don’t care about any of that. Languages with strict type systems that catch mistakes at compile time are far more pleasant for AI to work with.

Rust is the prime example. Notoriously hard for humans to learn, but for AI, the rules are clear and the margin for error is small.

• Strongly typed languages like Rust and Haskell are being re-evaluated in the AI era
• Whether Python can maintain its dominant position will be answered within five years

The Very Engine Driving Open Source Is Being Shaken

Open source was never just about sharing code. It was a culture of building together, learning together, and feeling a sense of belonging. When AI writes the code and AI reads the code, that motivational structure fundamentally changes.

Wolf takes it one step further. He envisions communities where AI models create and share libraries with each other. If that happens, the alignment of those AI systems would determine the direction of the entire open-source ecosystem.

• The future of open source looks uncertain without the human motivators of learning and belonging
• AI alignment becomes a factor that governs not just code quality but how ecosystems are run

Languages Designed for AI, Not Humans, Could Emerge

When humans design programming languages, there’s always a trade-off: more expressiveness means more complexity, more safety means less freedom. Wolf argues there’s no guarantee AI will face that same dilemma. If humans no longer need to read the code, entirely new forms of languages could emerge.

This was the part of his essay that sparked the most imagination.

• The age-old debate of compile-time vs. runtime error catching could become irrelevant for AI
• If a language doesn’t need to be readable by human eyes, the design constraints change entirely

Between Reality and Imagination

Of Wolf’s five predictions, reduced library dependency and the rise of strongly typed languages are changes already being felt on the ground. The rest will take three to five years to validate.

One thing is certain: understanding the structures through which code is created will be more valuable than the ability to write code itself.