postgrok

2026-10-23-youtube-lets-get-rusty-rust-perfect-language

Properties3
tagsrust, coding-agents, type-systems, youtube
typesource
statusseedling

5 min read

YouTube: Is Rust the perfect language for LLMs?

Source: YouTube video lx9Q66VNCWA, “Is Rust the perfect language for LLMs?” https://www.youtube.com/watch?v=lx9Q66VNCWA, uploaded 2026-10-23 on the Let’s Get Rusty channel https://www.youtube.com/@letsgetrusty. oEmbed record read 2026-10-25: title and author as above.

Ingested: 2026-10-25, from the auto-generated English captions of the full video (about 7.5 minutes), filed at Inbox/2026-10-23-is-rust-the-perfect-language-for-llms-lx9Q66VNCWA-transcript.txt. The presenter’s claims were checked against primaries before filing; the corrections and one surviving gap are recorded below and in # What held up and what did not.

Summary

The channel lets-get-rusty argues the case that greg-brockman made in a January 2026 post: “rust is a perfect language for agents, given that if it compiles it’s ~correct” https://x.com/gdb/status/2007228511363444905. The video stacks four industry events on top of the quote (bun’s Zig-to-Rust rewrite, openai’s Codex harness, Google’s C/C++-to-Rust campaign, the rea reverse-engineering tool), then borrows nicholas-matsakis’s framing for why the fit runs deeper than compilation: type-system guardrails the agent can be told to strengthen with newtypes and the type-state pattern. The turn of the argument deserves the record: Rust’s strengths (performance, reliability) are amplified by agents and its weakness (the learning curve) is minimized, because agents fight the borrow checker so developers do not have to. See rust-for-agents.

The video is also an advertisement. Roughly a third of its runtime promotes the channel’s “Rust Live Accelerator” course (1:52-2:13, 6:52-7:24), and the closing biographical paragraph is a career-sales pitch, not an argument. The claims survive that packaging, but the weighting should be read with it in mind.

The evidence pile

  1. Brockman’s quote (0:00-0:09): verified against the post itself, 2026-01-02, exact wording as quoted. See greg-brockman.
  2. Codex built in Rust (0:15): the presenter’s claim. Plausible and consistent with the OpenAI engineering line, but not independently checked in this ingest. Marked below.
  3. Bun (0:26-0:30): “Anthropic acquired Bun, and then the creator of Bun used Claude agents to rewrite Bun from Zig to Rust.” Both halves verified: Anthropic’s announcement 2025-12-03 https://www.anthropic.com/news/anthropic-acquires-bun-as-claude-code-reaches-usd1b-milestone, and Jarred Sumner’s own write-up 2026-07-08 https://bun.com/blog/bun-in-rust (about 535,000 lines of Zig ported in days with Claude agents, test-anchored). See bun, jarred-sumner.
  4. Google (0:37-0:49): “their new frontier model Argon… migrating C and C++ codebases to Rust across Google.” This sounded like a hallucination and is not: Google DeepMind announced Gemini 4 Argon 2026-09-30, and the announcement lists C/C++-to-Rust migration at scale (re2, libgav1, 800K+ lines of the Fuchsia Zircon kernel) https://9to5google.com/2026-09-30/gemini-4-argon-announcement/. See google-deepmind.
  5. REA (0:53-1:29): “somebody recently launched REA, which stands for reverse engineer anything… one developer took this tool and… re-implemented the entire Adobe suite in Rust.” Verified: morluto/rea https://github.com/morluto/rea, an MCP server wiring Ghidra, Hopper and IDA to coding agents, reported as the fastest-growing GitHub repo in history in October 2026, and the ArtCraft clean-room Adobe clones in Rust are covered at https://akitaonrails.com/en/2026-10-13/ai-reverse-engineering-is-adobe-dead. The “lawsuit to deal with” line is the presenter’s speculation. See rea.

Matsakis and the deeper fit

The video’s substance comes from nicholas-matsakis, co-lead of the Rust language design team, and his ISSTA 2026 piece “What Agentic Development Means for Rust (and Vice Versa)”. The captions garble his first name to “Nicholas Matsakis… Nico”, which resolves against his site https://smallcultfollowing.com/babysteps. Two of his points carry the argument:

  • Rust is a good substrate for building agents on: performance you can tune, a strong type system, a borrow checker that prevents memory bugs, and invariants the agent can be instructed to enforce with newtypes and the type-state pattern so invalid states cannot be represented (2:26-3:00).
  • Rust should keep getting more expressive for agents: low-level enough for a kernel, high-level enough for a GUI, with the compiler supporting more styles like GPU work (3:00-3:24).

The video’s own addition is the inversion of the JavaScript story (3:26-5:43): JavaScript ate the world because humans were the bottleneck and its ease-of-learning strength mattered. With agents writing the code, that strength stops mattering and its weakness (a loose type system that accepts wrong programs) is exposed exactly where an agent needs a machine to say no. See rust-for-agents and environment-and-constraints.

The closing caveat is the passage with the least sales pressure on it (6:01-6:17): Rust is not a silver bullet. You still need engineering judgment, ecosystem knowledge, and knowing when not to use agents. This is the same position Poteto’s software-factory argument reaches from the other side: the model is not the bottleneck, the environment and the reviewer are. See 2026-10-05-youtube-poteto-software-factory.

What held up and what did not

unsourced

“Microsoft recently used Copilot to rewrite the GitHub Copilot runtime from TypeScript to Rust” (0:32-0:36). The claim was not found in a primary source during this ingest and is recorded as the presenter’s assertion only. Every other claim the argument depends on was checked against a primary.

Sources

Graph View

Backlinks