Why a HashiCorp Founder Says AI Made Him Re‑Love Go

Mitchell Hashimoto, co‑founder of HashiCorp, famously declared that "Go has no place anymore" but recently posted a 180‑degree apology on X that attracted over 210 000 reads. He admits he was wrong because AI agents achieve astonishing productivity when working with Go.

“I started writing Go again… ‘Wait, didn’t you say Go has no place?’ I was wrong.” “The reason I was wrong is that AI agents are incredibly productive on Go. I won’t bring other languages into this because I don’t want to feed those crabs (a jab at the Rust community).”

Hashimoto highlights a paradox: CLI tools such as go doc and gopls have "shitty ergonomics" for human developers, yet they are a perfect fit for agents. An agent can invoke the LSP interface provided by gopls with minimal token cost to locate API implementations, method definitions, and call graphs.

“We always complained about Go’s verbosity, but that verbosity is exactly what LLMs love. The tools give them enough context to write runnable code on the first try.”

The article argues that Go’s extreme uniformity—enforced gofmt, a single for loop construct, and a minimalistic syntax—creates a deterministic surface for probability‑based AI agents. The agents do not need to guess whether a project follows a functional or object‑oriented style, nor handle complex ownership semantics; they simply follow the "Go Way" and generate code that works in the vast majority of cases.

Go + Zig: A "golden duo" for infrastructure

When a portable, low‑level library is required, Hashimoto prefers Zig over Go. He writes a simple directive in AGENTS.md: “Want an API or caller? Use gopls.” By leveraging Zig’s cross‑compilation capabilities, the CGO layer becomes a thin glue, avoiding the typical compilation pain of native Go‑C interop.

Go : high‑concurrency business logic and network scheduling.

Zig : performance‑critical, zero‑dependency, C‑ABI compatible core components.

CGO : minimized via Zig’s cross‑compilation.

This combination enjoys Go’s rapid development cycle while extracting Zig’s low‑level performance, offering a pragmatic alternative to the "Go vs Rust" debate.

Wes McKinney’s parallel insight

Wes McKinney, creator of the Pandas library, posted a blog titled "From Human Ergonomics to Agent Ergonomics". He observes that AI agents render Python’s human‑centric advantages moot and lists three fatal drawbacks for Python in the agent era:

Slow compile‑test loop: agents iterate far faster than humans, making Python’s startup latency a penalty.

Painful software distribution: agents need self‑contained binaries, whereas Python drags a heavy interpreter.

Performance and memory limits: issues tolerable for human developers explode under 24‑hour high‑frequency agent workloads.

McKinney concludes that Go (and increasingly Rust) will dominate AI‑focused infrastructure because they provide painless static binaries, deterministic builds, and low resource consumption. He notes that Go’s compilation speed gives it an edge over Rust in high‑frequency agent iteration scenarios.

“I still love Python and its ecosystem, but clearly, with agents delivering massive productivity, I and most of the industry will write far less Python and embrace Go and other modern compiled languages.”

Takeaway

Both Hashimoto and McKinney illustrate that a language’s "simplicity, predictability, and machine‑friendliness" become the most valuable traits when AI agents generate code. Go’s design—minimal syntax, uniform tooling, and fast compilation—aligns perfectly with the needs of LLM‑driven development, turning previously‑criticized ergonomics into a strategic advantage.