Summary
How much doubt comes from a new tool, and how much comes from knowing AI wrote it? The hosts discuss how teams can test those concerns and decide whether a change is useful.
Judge the Tool and Its Maintainers
Matt introduces slop bias through his team's discussion of tsrs, a Rust rewrite of the Go-based TypeScript compiler. He asks whether people combine valid concerns about a new project with an assumption that AI output must be poor.
The hosts consider maintenance history, known maintainers, language knowledge, and security risks. Scott asks for a working trial and measured benefits. He also questions the double standard of rejecting AI-written dependencies while using AI to write application code.
Engineers Still Own the Decisions
An engineer must review the output, choose the architecture, and correct errors. The hosts discuss how detailed instructions and tests can help an agent complete a rewrite. Human-written software also had defects and technical debt before AI tools became common.
Make Room for Dependency Work
Dillon suggests that resistance also comes from workload. Teams want to finish product features, while dependency upgrades compete for time. Calling a tool slop can become a reason to avoid that work.
Matt argues that cheaper prototypes and broader test coverage should make teams more willing to try changes. Scott agrees that a small trial can provide useful evidence.
Start with a Small Trial
Dillon distinguishes build-time tools from software that runs in production. Matt describes testing tsrs in the builds for internal build tools. The first goal is to gain confidence before wider use. A small local speed improvement can also add up across many applications.
The hosts return to the previous episode's discussion of getting buy-in. They discuss when to stop an extended debate, make a decision, and use later changes to improve the result.
Design System Detour
Memories of oversized product card components interrupt the main discussion. The hosts recall the difficulty of making one component handle every possible use. They close with jokes about future compiler rewrites in C and Zig.