Is Go the Best Language for AI Agents?

getbruin.com posted "Go Is the Best Language for AI Agents" and it got 300+ comments on HN. Here are the core arguments and counterarguments from a Rubyist's perspective.

The Original Post's Core Argument

AI agents generate code at scale. Most of it merely "looks plausible," so verification is the key challenge.

Go is well-suited for this verification:

Compiled language advantage — Static typing catches wrong types/arguments at compile time. If compilation passes, you're guaranteed syntactically correct code. No more "discovering TypeError in production" like Ruby.

Go over Rust — Go's simpler syntax means LLMs generate more idiomatic code. Go compiles faster, shortening the agent's feedback loop. More Go code in training data.

Standardized tooling — gofmt, go test, go build are standard. No JS-style "this project uses Prettier, that one uses ESLint" chaos. Tell an agent to format code and it runs gofmt.

Cross-platform binaries — GOOS=linux go build and done. No Ruby-style "this gem doesn't work on Windows."

Parts That Sting for Rubyists

Ruby's weaknesses are precisely targeted.

The problem with Python and Ruby: 20 different ways to do the same thing. Net::HTTP vs Faraday vs HTTParty vs RestClient. Agents can't predict which library to use. Go has net/http for most cases.

Agents generate valid Go code on the first try about 95% of the time (author's experience). Ruby/Python errors are only found at runtime. This gap matters in agent workflows.

HN Counterarguments

"Java works for the same reasons" — True. More training data, stronger type system. The post not mentioning Java is slightly biased.

"Rust is better" — Once you pass the compiler, runtime errors are near-zero. But Rust changes fast, making it hard for LLMs to keep up with the latest APIs. Good point that Rust tests live in the same file, so agents update tests when modifying source.

"OCaml/Haskell are best" — Powerful type systems (including GADTs), pure functions, fast builds. Theoretically correct but training data is scarce.

"Python was better" — Someone who rewrote projects in multiple languages found Python worked best with Claude. Shorter code, easier to understand.

"Domain matters more than language" — The most pragmatic take. TypeScript for web, Python for data/ML, Go for CLI/infra. The "best language" depends on what the agent does.

What Should Rubyists Do

Ruby's disadvantages for agent code generation are real. Dynamic typing, framework fragmentation, runtime error dependence. These are structural weaknesses.

But rewrite your Rails app in Go? No. As the original post admits, "in 12 months we might rarely read code directly." Current language trade-offs may diminish as LLMs improve.

Pragmatic approach: Ruby for business logic, Go for performance-critical parts. Convert Sidekiq workers to Go, build CLI tools in Go. Many teams already do this.

The real lesson isn't "Go is best" but that compile-time verification + standardized tooling + simple syntax gain value in the agent era. Whether Ruby can evolve in this direction is the more important question for Rubyists.