POV: Your coding agent just tried to phone home High school student Jake Burwick built Vow, an open-source MIT-licensed programming language designed to contain coding agents by requiring every function to explicitly declare its use of network, filesystem, or environment access. The vow --sandbox runtime inspects these capability declarations before execution and denies any program that reaches for undeclared resources; in a November 2026 test, Vow blocked seven out of seven unauthorized attack programs. The language, roughly 6,400 lines of JavaScript, also removes nulls and exceptions in favor of Result values and formats errors as repair instructions for agents. You stare at the terminal. You asked the agent to parse a CSV file. It wrote the code, it ran the code, and then it tried to open a socket to a server in a region you have never visited. Coding agents run on a loop. They write, they run, they read the error, they fix. Most languages treat agents like human developers. They assume the code is benign. Vow treats agents like the machines they are. I am Jake Burwick, a high school student, and I built Vow to keep that loop contained. Vow is open source under the MIT license, with the source code at github.com/JakeBurwick/vow. Every function in Vow must explicitly declare if it uses net, uses fs, or uses env. If the code reaches for the network or your file system without a declaration, it fails. The command vow --sandbox inspects these declarations before a single line runs. If an agent tries to touch your environment variables without permission, the runtime stops it cold. bash $ vow --sandbox --allow fs=./data app.vow x SandboxError: needs capabilities the sandbox does not grant env: fn leak declares 'uses env' Nothing was run. The runtime denies the program entirely. This is not a suggestion. It is a hard lock. In a November 2026 test, Vow blocked seven out of seven unauthorized attack programs. Vow also eliminates the silent failures that usually keep agents stuck in their loop. It has no nulls and no exceptions. Functions return Result values that you must handle with match or the ? operator. If the agent ignores a failure, the code refuses to build. When an error does happen, Vow formats the output as a repair instruction. The agent does not have to guess what is wrong. php x TypeError: can't add a num and a str - line 6, col 10 6 | return "Total: " + sum ^^^^^^^^^^^^^^^ Likely fix: convert the number to text first: str sum The entire language is about 6,400 lines of JavaScript. It is fast. A recursive fib 32 runs in 0.09 seconds, while Python takes 0.14 seconds, measured in October 2026. You can install it with npm: npm install -g github:JakeBurwick/vow If you use Claude Code, you can add the MCP server to give your agent the tools to guide, check, and run Vow code. claude mcp add vow -- vow-mcp You can paste a one-paragraph prompt from the site into any coding agent to point it at the guide. The site also has an llms.txt and an AGENTS.md file you can drop into your project. Explore the playground, tutorial, and docs at https://vow-lang.vercel.app https://vow-lang.vercel.app The next time your agent writes a function that tries to read your environment, the sandbox will stop it before it even starts. You can go back to your coffee.