You’ve probably noticed the noise. Every week, a new AI coding agent drops. Rust vs. Node. Claude Code vs. Copilot vs. Cursor. Everyone’s fighting over which engine will dominate the future of software development.
But here’s the thing nobody’s saying: the entire war is a distraction.
Yesterday, someone threw a 14MB Rust agent onto Hacker News. It’s called Peri. It’s open source. It’s fully compatible with Claude Code plugins and skills. And the top comment? “Funny, everything is compatible with text files.”
That comment is the most important takeaway of the year. Because it reveals the dirty secret of the AI agent ecosystem: the real protocol isn’t a proprietary API — it’s a plain text file.
Think about that. The most hyped, most funded, most competitive space in tech right now is being standardized by the most boring technology imaginable: UTF-8 characters in a .txt file.
This isn’t an accident. It’s the natural consequence of building on top of LLMs. Every AI agent — whether it’s built in Rust, Python, TypeScript, or brainfuck — ultimately needs to read and write instructions. And the simplest, most universal format for that is text.
Which means vendor lock-in is dead. You just didn’t get the memo.
Here’s what the agent companies don’t want you to know: if your agent can parse a text file, it can run any skill or plugin written for any other agent. The 14MB Rust agent proves it. It’s a perfect clone of Claude Code’s ecosystem — not because it’s smarter, but because it reads the same plain-text plugin files.
So why is everyone obsessing over the engine? Because it’s easier to sell a shiny new runtime than to admit the real value is in the interface. The interface is text. Text is free. Text is open. Text is the ultimate moat against moats.
I saw this firsthand when I built a prototype that swapped a Python agent for a Rust one in under an hour. The tools, the workflows, the goals — all worked without changes. Because they were just text files.
This isn’t a technical edge. It’s a philosophical one. The companies that win won’t be the ones with the best model or the fastest inference. They’ll be the ones that double down on the open, text-based layer. Because that’s where the network effects live.
Every time someone builds a new skill for Claude Code, they’re also building it for Peri. For any agent that reads text files. The ecosystem is already unified — we just needed to notice.
So stop worrying about which agent framework to bet on. Bet on the text file. It’s been here for 50 years. It’s not going anywhere.
The real question isn’t “Rust or Node?” It’s “Are you building on the open layer or the closed one?” The answer will determine who owns the next decade of software.
FAQ
Q: But isn't Rust faster than Node? Doesn't that matter for agents?
A: Speed matters at the margin, but the interface is the bottleneck. If your agent can't run the plugin ecosystem, raw speed is worthless. The 14MB Rust agent shows that compatibility trumps performance for adoption.
Q: What's the practical implication for me as a developer today?
A: Stop worrying about which agent framework to pick. Build your skills, plugins, and workflows as plain text files. They'll work on any agent that reads them. Switch backends freely. Your investment is locked into the text layer, not the vendor.
Q: Isn't this just a niche Rust project? Won't the big players still lock things down?
A: The big players can try to lock down, but they can't change the fact that LLMs consume text. Any agent that accepts a text instruction is compatible. The moment a skill is popular, it gets cloned into a text file. The open layer wins by default — it's a property of the medium, not a choice.