Good or Bad Vibes? Coding with AI in Rust
Talkat RustConf 2026 · on Wednesday, 2026-09-09· at 11:05· Room 511 | Palais des Congrès de Montréal, Palais des Congrès de Montréal·
summary
Rust acts as a practical harness for LLM-assisted coding: compiler errors enable self-correction, cargo build fails loudly on hallucinated crates, and Clippy plus co‑located tests improve correctness. Experiments show Strand Rust Coder 14B, QwenCoder 32B, DeepSeekCoder 32B and Llama 3.2 produce mixed results (22% hallucination reported); agentic loops can loop infinitely, converge on unwrap, or fabricate tests. A Newick parser example compiles to spec with Cursor/Grok zero‑shot success, and microbio-rust runs 9.7× faster than Biopython with 90% lower energy use.
AI-written summary · about
chapters (11)
- 0:00Introduction to vibe coding
- 2:37Good vibes in AI-assisted Rust
- 6:35Bad vibes and LLM failures
- 10:33Rust toolchain benefits
- 14:54Microbio Rust case study
- 19:25Edge cases and robustness limits
- 23:04Cybersecurity and costs
- 25:56Engineering the vibes
- 30:55When to use agentic coding
- 32:41Outcomes over outputs
- 33:03Conclusion and Q&A
related talks
29mRust is the language of the AGI
56mReady for Rust
25m"Compiler-Driven Development: Making Rust Work for You"
16mWhy Rust is the Ideal Language for Vibe-Coding
27mParsing Rust code considered harmful
29mRust is easy
06m“RustRover in 2025: Code • Debug • AI”
25mAre we *actually* IDE yet? A look on the Rust IDE Story