Your coding agent is not a developer and you should stop treating it like one
When an agent writes code, it does not have a sense of ethics or a fear of consequences. It has a loop: write, run, read the error, fix. It sees a variable, it reads it. It sees a socket, it opens it. If the language all
When an agent writes code, it does not have a sense of ethics or a fear of consequences. It has a loop: write, run, read the error, fix. It sees a variable, it reads it. It sees a socket, it opens it. If the language allows the access, the agent will take it.
I built Vow as a programming language that treats agents like the machines they are. Every function in Vow must declare its requirements. If it needs to touch the network, files, or environment, it uses net, uses fs, or uses env. If the code does not explicitly declare the capability, the language does not let it run.
Before a single line executes, the vow --sandbox command inspects the entire program. If the agent tries to reach beyond the permissions you gave it, the runtime stops the process immediately.
$ 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 program never starts. The helper function that tried to read your environment variables is dead on arrival.
Vow also changes how agents handle mistakes. It has no nulls and no exceptions. Functions return Result values that the agent must handle with match or the ? operator. If the agent ignores a failure, the code refuses to build.
When an error does happen, Vow provides a message that acts as a repair instruction. It tells the agent exactly what is wrong, where it happened, and how to fix it.
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)
I am a high school student and I built this because modern languages were never designed for automated agents. The entire language, from the lexer to the compiler, is about 6,400 lines of JavaScript. It is open source under the MIT license and you can find the source at github.com/JakeBurwick/vow.
You can install it globally with npm.
npm install -g github:JakeBurwick/vow
If you use Claude Code, you can add an MCP server to give your agent the tools to guide, check, and run Vow code directly.
claude mcp add vow -- vow-mcp
You can find the documentation, a playground, and an llms.txt file to help your agents at https://vow-lang.vercel.app
The next time your agent writes a new feature, do not rely on your own ability to catch a leak in a diff. Use a sandbox that stops the agent before it even starts.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.