Ray Myers wrote up a response to a response of a... oh, heck. Here's the link: "Zig Creator Calls Spade a Spade, Anthropic Blows Smoke." He's considering Andrew Kelley's response to Bun being rewritten in Rust, and editorializing somewhat along the way. It's justified, but the emotion he writes with leaks into his commentary.
It's an interesting post that covers a LOT of ground:
- What happened to start all of this (Bun's team used Anthropic's AI tools to rewrite Bun from Zigto Rust, after Anthropicacquired the project) - Zig author Andrew Kelley's analysis of the migration, which he published in two passes, one more emotional than the other, along with some personal observations
- Some interesting aspects of the entire project: market positioning, differing claims on everyone's part, and some questions about the use of AI and how it's applied that we can take away for our own projects
The bottom line is that the marketing aspect weighs very heavily on everyone here, including the downstream audience: the engineers and executives who will make real architecture and staffing calls off the back of this story.
The technical achievement is pretty strong - with "pretty" being important here, because one would hope, as Myers observes, that a "great port" would have left fewer holes, and some of the restrictions on the agents themselves show cracks in process: the marketing implies the humans were incidental, so why did the agents need so many human-written guardrails to stay on task?
Something doesn't add up here, and it comes down to Anthropic itself: the port is not just technical, but is being used as a marketing exercise. It works, and has merit, but it's also remarkably difficult to analyze properly, because the codebase is too large and too much a port of code from a different language (while retaining the structure of that language's code). And if the agents really are all that and a bag of chips, the porting instructions shouldn't have had to include some of the notes that turned out to be necessary.
Yet there's also some good advice about agentic work: the porting team set the agent that wrote a change against separate reviewer agents, each in its own context window, providing diffs as input with the assertion that the diffs contained code that was broken, a fascinating approach to stress-testing code using an agent's perspective.
That's the real takeaway - discarding the emotions present on all sides, this is a worthwhile technical event dressed up in marketing clothes, and that's where most of the friction occurs: everyone wants to look their best in the public forum.