Most programming languages were built for humans. You write the code, you read the error messages, and you fix the problems. That assumption runs deep — and it’s starting to create friction as AI agents take on more of the development workload.
Vercel Labs is taking a direct shot at that problem. The company recently introduced Zero, an experimental systems programming language designed from the ground up with agents as the primary users. The project, which was built by Vercel engineer Chris Tate, asks a simple question: What does a programming language look like when machines need to use it as much as people do?
A Language That Agents Can Actually Use
Most programming languages were designed for humans who read error messages, interpret warnings, and manually trace through stack output to fix bugs. AI agents don’t do any of those things well. They work better with structured data: Predictable tokens, stable codes, and machine-parseable repair hints.
Zero is built around that reality. The compiler emits JSON diagnostics with stable codes and typed repair metadata, so agents can parse and act on errors without interpreting human-readable text. Commands like zero explain and zero fix –plan –json give agents structured access to diagnostic explanations and machine-readable fix plans — no prose parsing required.
That’s a meaningful shift. When an agent encounters a compiler error today, it has to interpret a message written for humans. Zero skips that translation step entirely.
Explicit by Design
Zero is built around “everything is explicit” — no hidden allocators, no implicit async, no magic global variables. Effects are declared in function signatures. The compiler enforces them at compile time. Memory and control flow stay predictable throughout the program.
That explicitness matters for agents. When behavior is hidden or implied, agents have to guess. Zero prefers one obvious way to express most things, even when that makes code more explicit than a human might choose in another language. Regularity over syntax, as the project puts it.
“Developer tooling is being redesigned to serve agents as first-class consumers. Zero treats compiler diagnostics, repair plans, and dependency facts as structured data agents can act on directly, opening the compiler layer to two consumers in parallel,” said Mitch Ashley, VP and practice lead, software lifecycle engineering, The Futurum Group.
Zero sits in the same design space as C or Rust, but its design priorities differ. It’s not optimizing for developer ergonomics first. It’s optimizing for agent reliability — and then making that usable for developers too.
Small Footprint, Fast Output
Zero compiles to sub-10 KiB native binaries, which keeps the output lean and deployable across different environments. The toolchain includes structured diagnostics, dependency graph facts, binary size reports, and structured fix plans — all of it consumable by agents without additional parsing or interpretation.
Vercel CTO Malte Ubl shared an experiment showing that a Bun rewrite using Zero achieved a 98.7% test pass rate within roughly 22 hours while significantly improving compile times. That benchmark hasn’t been independently verified, but it has generated real discussion across developer communities that are paying attention to AI-native tooling.
Early Days, But Real Traction
The GitHub repository crossed roughly 1,500 stars shortly after launch, with v0.1.1 listed as the latest release as of mid-May 2026. The project is pre-1.0 and explicitly unstable — the team is making breaking changes as it searches for the language, library, and tooling patterns that work best for agents.
The repo’s documentation puts it plainly: Treat today’s syntax and APIs as something to explore, not something to memorize.
Zero isn’t the only thing Vercel has been building in the agent space. In January 2026, Vercel launched skills.sh — described as “npm for AI agents” — letting developers install best-practice skill packs for Claude Code, Cursor, and other tools. It reached more than 20,000 installations within hours. Zero extends that direction at the language level.
What Comes Next
“Toolchain vendors face a real question about whether existing diagnostic surfaces can support agent workflows, or whether agent-native alternatives will displace them at the language layer. Teams shipping the most agent-generated code will hit that decision first,” Ashley added.
Zero is not production-ready. The project explicitly warns against using it with sensitive data or trusted infrastructure. But it represents something worth watching: A deliberate attempt to rethink the programming-language contract for a world in which agents write, run, and repair code alongside developers.
Programming languages are evolving from “human-only” to “human-machine shared.” That doesn’t mean humans no longer need to understand code — it means code and toolchain output must serve two consumers simultaneously: intuitive, creative humans and fast, tireless agents.
Zero is an early experiment in what that looks like. Whether it takes hold or gets absorbed into something else, the problem it’s trying to solve is real — and it’s only going to get more pressing as agents become a bigger part of how software gets built.
