DESCRIPTION Richard Feldman and Anjana Vakil trace the Roc programming language's ambitious ground-up compiler rewrite — from Rust to Zig — which happened to coincide almost exactly with the year AI coding assistants went from useful to transformative. Richard describes how AI's role shifted over just 12 months: from mechanical test-porting grunt work, where it was reliable but limited, to genuine architectural collaboration on harder problems. The key insight is that guardrails don't live in prompts ("never do this" gets ignored constantly), they live in the code itself — invariants and automated feedback loops that catch the AI when it strays, rather than instructions it will cheerfully disregard.
The conversation widens into what the AI era means for software quality and trust. Both are wary of the coming wave of AI-generated "slop" — buggy, mediocre software produced at scale — but Richard makes the counterintuitive case that competitive pressure might actually force quality up: if everything is slop, the products that aren't will stand out hard. Anjana draws a parallel to consumer electronics brand trust: just like we pay a premium for the USB-C cable we know won't cause a fire, developers and users will increasingly gravitate toward names and communities they can vouch for. The open source contribution model, they agree, needs new systems to navigate this — and Roc v1, due before the end of 2026, will be a test case.
RECOMMENDED BOOKS Richard Feldman • Elm in Action • https://amzn.to/387kujI Dean Bocker • Don't Panic! I'm A Professional Zig Programmer • https://amzn.to/3ljKT8d Tim McNamara • Rust in Action • https://amzn.to/3ux2R9u David Drysdal • Effective Rust • https://amzn.to/4dAjbdX Eric Normand • Grokking Simplicity • https://amzn.to/3gz7o3C
Podden och tillhörande omslagsbild på den här sidan tillhör
GOTO. Innehållet i podden är skapat av GOTO och inte av,
eller tillsammans med, Poddtoppen.
