@charliermarsh: Talked with @ryanlpeterman about how software engineering is changing! I like the pull quote, which captures a lot of w…

Talked with @ryanlpeterman about how software engineering is changing!

I like the pull quote, which captures a lot of what I’m struggling with now: “Does it even matter if we understand the code?”

https://t.co/GpSEf1pgYa

TL;DR: Charlie Marsh discusses the transformation in software engineering: AI-written code makes PR review cost constant while generation cost drops to zero. He shares his experience building Ruff, a Python tool written in Rust from scratch, emphasizing the importance of fast iteration, authentic marketing, and performance benchmarking.

Software engineering is undergoing a massive shift

“It’s a really interesting time to build software because everything is changing so fast.” When someone submits a PR, Charlie Marsh believes the review needs extra scrutiny: “It’s not you writing the code anymore — it’s an agent. The cost of submitting a plausible PR has dropped to zero, but the cost of reviewing it stays the same.” He worries about junior engineers entering the field: “If they rely so heavily on agents to build, how much do they need to understand the different parts of the codebase?”

From computational biology to developer tools

Charlie is the founder of Astral (the Python developer tools startup later acquired by OpenAI). Before starting Astral, he was the second engineer at a computational biology company. With no background in biology or machine learning, he was responsible for building all the software systems used for ML, primarily in Python, with some Go and Rust. He noticed that the Python ecosystem lacked the experimental spirit of other ecosystems (like the web). “At the time, a lot of things seemed pretty crazy, and there was a stronger focus on performance — native JavaScript tools like ESB (in Go), SWC (in Rust), Bun, Deno were becoming popular, but that wasn’t happening in Python.” His key question was: “Why can’t we have these?”

Building Ruff: starting with a linter

Charlie’s exploration began with a blog post titled “Python tooling can be much faster.” He built a linter prototype first because “building a linter is indeed much easier than building a type checker.” A linter has a simple core but a rich set of rules, allowing users to get value quickly and iterate. In contrast, “a type checker that’s 75% complete is not useful.” By shipping a usable tool fast, he built momentum: “If you can confirm a problem, fix it, and ship a version within a day, that’s a very powerful loop, and you’ll gain supporters.”

Launch reception and marketing philosophy

That blog post hit the Hacker News front page, generating excitement and interest. Charlie emphasizes the “attention economy”: people have 10 seconds to decide if they care about your project. He avoids clickbait or misleading claims, and instead “tells them honestly and sincerely why they should care.” He gives two examples:

• Benchmark chart: Ruff’s benchmark chart spread on Twitter, visually showing the performance difference. “All benchmarks are flawed, but a visual hook can quickly explain why you should care.”
• Tagline: “Compatible with existing tools, but much faster” — simple and direct, conveying the core value.

He also notes that “developer marketing” can sound like a dirty word, but “there are thousands of genuinely good projects on GitHub that never get discovered or grow because they don’t know how to market themselves.” He believes you need to seriously think about how to quickly communicate importance.

Why Rust

When asked why he chose Rust, Charlie admits: “A large part is Rust’s hype… I saw Rust everywhere, it was growing fast, and people said it seemed like an easy way to write high-performance software.” In hindsight, he thinks it was the right choice.

Source: @charliermarsh: Talked with @ryanlpeterman about how software engineering is changing! I like the pull quote, which captures a lot of w… (https://www.youtube.com/watch?v=Iw65FD4MGgs)