Sure looks like Bun is at least exploring a port from Zig to Rust given this docs/PORTING.md guide for coding agents github.com/oven-sh/bun/bl…

Confirmed that this is an experimental branch, not a decision to switch languages

@simonw more slop in these replies than the code

@simonw This is obviously the right decision, and I might go as far to say that Zig was a mistake

Ready to bring ideas to life? Try Grok Imagine today

@simonw it would make more sense writing a borrow checker for Zig

@simonw It would be so funny if they converted their Zig compiler fork to Rust but continued to use _that_ Zig compiler for Bun

@simonw Half the Zig to Rust pull right now is the AI coverage gap. Models know Rust well. Zig has way less training data. Pick a niche language and your agent dev slows down. Its a new variable in the choice now.

@simonw If you can't upstream important patches, and you don't want to maintain your own fork, this seems the better move.

@simonw Zig's comptime was the whole pitch, swapping to Rust is basically admitting the agent tooling moat won

@simonw Say it with me: The Machine Cannot Be The Final Judge of the Machine. - Carpenter of Impossible Systems

@simonw Nothing good will came out from it. Thankfully I removed every bun and bunx reference from my codebase. Bun and anthropic are truly supply chain risks.

@simonw Rust is not the best language to write stuff with models tbh, its very much designed for human verification with compiler screaming at you in every corner and models cheat often.

42 agents. 216 threads. One dashboard. Every agent gets its own prompt, tools, skills, and budget. Deploy specialized agents across your company. From the team at Airtable.

@simonw Seems like bun is now playground for slop ai projects, that's a pitty

@simonw the porting.md is the tell. zig has way thinner training-data weight than rust, so coding agents misalign on idioms constantly. cheap mitigation while you're still on zig: pin a 'use rust patterns when uncertain' note in the system prompt and the false-edit rate drops fast.

@simonw porting.md written for coding agents is the part most people skip in this story. migration guide targeting llm not human already tell you who doing the actual work on a codebase that size

@simonw Basically @deno_land but WebKit. Eventually it will land up on v8 too. Maybe stability outweighs speed is the whole takeaway.

This is the real shift. You trade Zig's 'every compiler feature is fair game' culture for Rust's 'every model has consensus on this pattern' culture. What disappears: the weird, performance-critical bits nobody's ML training data touched. Your runtime becomes more predictable and more standard. Not always a win.

@simonw a porting.md tuned for coding agents is the project contract. without it each file gets translated by local context and type-mapping gets reinvented per chunk. read the contract every turn or idiom drift compounds across hundreds of files.

@simonw It's kind of crazy how the skill files or other instructions that are ending up in these projects now are showing intent in ways that the code itself never could have told you.

@simonw Thank you so much for posting. I am learning Zig currently and being experienced in Rust, this will be very helpful to map concepts. Also a good learning experience on how to interact with agents.

Take advantage of your New Rewards Member Offer today!

@simonw What's the read on whether the agents-as-porters thing actually scales here, or is the doc more aspirational than load-bearing?

@simonw zig to rust pipeline would actually be a massive undertaking though

@simonw the real question is whether coding agents can meaningfully port Zig->Rust without introducing memory safety regressions in the translation. PORTING.md is a nice signal but the unsafe blocks will be the hard part

@simonw wait, bun is adding porting docs specifically for coding agents? that's actually huge if they're thinking about agent workflows from the start rather than bolting it on later

@simonw Rust has a larger community, more mature tooling, an easier hiring funnel. Interesting experiment

@simonw the PORTING.md is the interesting part. they wrote documentation for the agent, not for humans. 'Phase A captures logic but doesn't need to compile' is exactly how you'd decompose this for an LLM - semantics first, build errors second. that's a new genre of documentation

@simonw interesting that Rust won even here, zig was supposed to be the systems play

@simonw Porting docs are usually where the real pain shows up, not the headline. If Bun is moving from Zig to Rust, the interesting part is how they handle tooling and long-term maintainability. I’ve had similar ‘stability first’ moments with Hyperagent.

@simonw wait they're actually considering rewriting the entire runtime? that's either brilliant or absolutely insane lol

THE BRUNSON BURNER HAS IGNITED! 14 points (6/7 shooting)... in 9 minutes 😮

@simonw I wonder how this performs in practice with SO MANY independent instructions.

@simonw Looking at the porting guide tells me this is a terrible idea. I don't think the Zig to Rust file by file translation will add any meaningful value. It feels lazy and downright stulid. Rather layout all the systems and do a bottom up migration and multiple phases after that.

@simonw Is that related to Zig rejecting the Bun Zig fork changes?

@simonw PORTING.md as the contract is the right move. Most migrations fail because tribal knowledge never made it into the repo. Whether that file holds the actual invariants is the whole game. How is Bun verifying behavior matches after a pass?

@simonw repo → route ≠ owner “PORTING.md” → agent execution lane compile ≠ behavior translation ≠ equivalence PR ≠ permission merge ≠ control self-directed rewrite ≠ self-certified safety named reviewer + denial gate + rollback path = external brake

@simonw The real bottleneck isn't reasoning, it's context window pollution. Even Claude 3.5 Sonnet degrades when >50% of the 200k window is noise. SWE-bench Verified shows most agents fail on repo-level navigation, not code gen. We need better retrieval, not bigger models.

@simonw interesting porting strategy — phase A for semantic correctness before even worrying about compilation is smart. rust's ownership model will catch a lot of the memory issues zig leaves to the programmer. curious if they're doing full migration or keeping zig for any hot paths 🦀

From a Pro 200 cron user: this porting doc means more than 'just experimental.' Anthropic is testing whether the coding-agent ecosystem can actually port Bun's Zig surface into idiomatic Rust under guidance — that's the real bet. If agents nail this, expect more performance-critical ports getting framed as 'experimental' first.