What am I talking about? Nil and partially constructed structs are impossible to prevent the creation of in Go.
Sure, if you’ve got a small program with limited scope, that’s probably fine if you look through squinted eyes. But the teams I work with are working on sprawling, evolving software where the compiler saying “hey, that’s not a valid Widget” would be extremely useful and save much heartache.
An LLM does a good job of “checking” for other uses and “checking” if everything is going to work correctly, but - supposedly we’ve committed the concept to code so that the compiler can actually verify it - and Go intentionally permits invalid states of structs. This makes Go a fundamentally problematic language choice for the kind of software I work with teams on, LLM or not.
But on the other hand, I suppose it makes it a bit more pragmatic - less checks makes for a faster compiler, and fast compilation was/is very high up in the language's requirements and motivation. If you want / need more strictness in your language, there's Rust, Java, C#, etc.
It's sometimes challenging to get a Rust program to compile... but if you do, it's probably going to work.
As for readability, the fact that AI-written Go closely resembles human-written Go is not necessarily a point in Go’s favour.
There's just not many ways of writing Go. It's a very dull language. It was designed to be dull and easily understandable.
if an llm has learned dumb things from dumb users it could disproportionately cause provlems versus other languages, just by being "in a sloppy mood" when writing go.
For me it helped a lot to try to make containerized end-to-end tests and a custom TestMain for this, where I am using podman to run the integration tests. This way the end-to-end tests are forced to be on network level, and you can test protocol and API quirks much easier with LLMs.
Also, never forget to write a bootstrapping docs/ folder so that you don't have to re-explain these things all the time.
In Go the convention is kind of to have a constructor pattern with a NewStruct(...) *Struct method that initializes all properties.
Also can't you build your own validator for that with the reflect package in the Add() method of your UI graph to prevent this sorta thing?
But that doesn't stop you from declaring a var s Struct, and never initializing it, or making a NewStruct {}.
> can't you build your own validator for that with the reflect package in the Add() method of your UI graph
Besides the fact that that would almost certainly significantly hurt performance, how would you be able to differentiate between unitialized data and data that was intentionally set to the zero value?
The go type-system is simply incapable of enforcing nil-safety without being no longer able to compile the go stdlib nor most code in the wild, so it's a quite valid criticism of the go type-system and language, and your comment doesn't hit on a valid solution.
At Netflix, I lead the Go language guild. We've been seen increasing reports of users finding their AI agents writing better Go code than other languages, and increasing reports of projects favouring Go over other languages.
Two additional notes I'll add:
- Go has _great_ resources on writing good Go code, including treasure troves at https://go.dev/doc/effective_go and https://google.github.io/styleguide/go/. edit: Sorry, I forgot to add: we give these resources to AI agents and they use them to produce even better Go code.
- For a language team, Go is a dream. The `go fix` tooling, AST/SSA packages, ease of reading and writing `go.mod` (go mod edit, etc), and various other "platform"-y features make modifying Go code at scale way easier than other languages.
As another example, Go still has not yielded a correct implementation of Raft or Paxos while there are dozens in Java, C++, and Rust. Antithesis found some more bugs in HashiCorp's Raft implementation recently[0]. I'm sure etcd still has some kicking around.
Maybe this is a "don't throw the baby out with the bath water' problem but the general evolution of Go has been lackluster. I reach for Rust, Zig, and modern Java instead depending on the specific needs and constraints.
0 - https://antithesis.com/blog/2026/finding-bugs-in-raft-implem...
Are you saying that this implementation is wrong?
"This Raft library is stable and feature complete. As of 2016, it is the most widely used Raft library in production, serving tens of thousands clusters each day. It powers distributed systems such as etcd, Kubernetes, Docker Swarm, Cloud Foundry Diego, CockroachDB, TiDB, Project Calico, Flannel, Hyperledger and more."
One of the most popular distributed DB is Cockroach which is written in go and also uses Raft: https://github.com/cockroachdb/cockroach/tree/master/pkg/raf...
You may be interested in knowing that the largest managed Kubernetes service in the world (AWS EKS) ripped out etcd for in favor of their homegrown consensus service for large scale EKS clusters: https://aws.amazon.com/blogs/containers/under-the-hood-amazo...
goBGP is arguably even worse.
I don't have a third place in mind that's even worth mentioning relative to these two.
(I have only a rather basic familiarity with go, but was considering gobgp for an infra project...)
[1] https://elegantnetwork.github.io/posts/comparing-open-source...
Gobgp is great if you want to embed it directly into a Go app though. Talos Linux has done that recently.
0: https://etcd.io/blog/2025/autonomus_testing_with_antithesis/
Etcd corruption and loss of quorum is extremely common in practice and the GitHub issues sit for years. The design is simple, the performance is modest, yet it still has still never been reliable, despite being marketed as so. I can't speak to whether this is specifically due to their Raft implementation, but I'd argue the entire codebase is over-engineered and questionable.
Its very much {reliable, performant, flexible} pick none.
That the world runs on Kubernetes is no qualitative statement about the correctness of its Raft implementation. You can say that it's clearly good enough to not matter most of the time, but that is a different statement. No matter who you look at, they're just cooking with gas like you do, and they can make mistakes in just the same way.
Now; I'm only attacking your argument. I do neither know nor particularly care about the correctness of that implementation itself. There's been better refutations of the claim you replied to in other answers anyway.
Thats basically it for starters, what non-technical solution do you propose?
But k8s was invented so that "dev" and "ops" teams could play nicely when they are different org structures, with their own budgets, deadlines, etc.
If it solves some technical issue then that's incidental.
I didn't know Go just isn't a good language for it, but now that I know I'm no longer surprised at etcd being problematic.
Sure it may not be the best fit in a scenario where you want a cluster spanned over the entire globe (thats why GKE uses paxos-based Spanner instead of it) , but even spanned across an entire continent (in europe via glass fiber) it works quite well for me. Its one of the least problematic parts of the stack.
> are you saying this implementation is wrong?
> That's not remotely what he's saying at all.
I'm v confused by this thread
That is literally what the comment says.
I hope my every competitor will take your advice to heart, as one of our competitors did when they read that "Go is not a memory safe language", so they wrote a blog about how they are porting to Rust. While our team was moving fast and using those "primitives that should almost never be used" around our long running production code base with success.
Some time has passed and now their company does not exist anymore and we have a lot of their clients.
Thank you!
> Go's internal data structures like interface values, slice headers, hash tables, and string headers are not immune to data races, so type and memory safety can be violated in multithreaded programs that modify shared instances of those types without synchronization.[113][114]
It's close enough for most purposes... but it isn't.
JS is fine for what and where it is, Rust is fine too. I just appreciate the stupid simple nature of Go and it does the job just fine.
If you want to see something that could potentially impact Raft's correctness, search the last couple of days of the HN front page for choreographic languages [1]. But none of these are even remotely mainstream enough to depend on for anything. Nor do I know if anyone in these languages has implemented Raft. A rather good test case for them, if any of them are looking. That's something that could actually help a Raft implementation's correctness, not just fiddle around the edges of local concurrency issues.
[1]: https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...
The antithesis author states:
"we’ve found bugs in every Raft implementation we’ve tested, including HashiCorp Raft, Aeron Cluster, OpenRaft, and MicroRaft"> Go still has not yielded a correct implementation of Raft or Paxos while there are dozens in Java, C++, and Rust.
That says that there are correct (i.e., bug-free) implementations in those languages. The GP noted
> "we’ve found bugs in every Raft implementation we’ve tested, ..."
which says that there aren't any correct ones. You then wrote
> I didn't say other languages don't have buggy Raft/Paxos implementations
which is a strawman. The issue is whether there are correct implementations. That there are buggy ones is irrelevant.
(FWIW I have no dog in this fight ... I'm just reading here.)
>> "we’ve found bugs in every Raft implementation we’ve tested, ..."
> which says that there aren't any correct ones
That only follows if the GP tested every Raft implementation in existence and no new ones were written since.
Edit:
> What are they implying by citing that? That Raft implementations in all languages have bugs?
That's what it says.
> I've already pointed out that is false.
You claimed that, and it's being disputed.
> Please let me know, since you're so comfortable speaking for them.
This has veered into bad faith ... I won't comment further.
> The antithesis author states:
> "we’ve found bugs in every Raft implementation we’ve tested, including HashiCorp Raft, Aeron Cluster, OpenRaft, and MicroRaft"
What are they implying by citing that? That every language has a Raft implementation with bugs? Yes that's probably accurate because lots of people make Raft implementations for fun and learning. Again, Go does not have a single Raft/Paxos implementation that is rock solid. I have seen many in C++, Java, and Rust that are doing tens of millions of requests per second in production for over a decade.
Is their point that Go is not the only language with this problem? My post already points out the track record is that Go is the problem for writing correct code in highly critical domains.
The only way this becomes useful for comparing languages is if somebody gives evidence of correct implementations in other languages. You're claiming they exist but with no evidence and suggesting they're secret. How do you know those don't have bugs? Did any concurrency bug experts do extensive testing on them? And can we disprove secret Go implementations of the same quality?
No, they are citing that every Raft implementation that Antithesis has tested has bugs. The etcd implementation you note in go that has bugs also does tens of millions of QPS and is over a decade old. How are you confident that the proprietary implementations that presumably haven't been fully tested don't have subtle bugs that don't show up in practice?
But good god, the second it gets to anything concurrency-related, it just loses its mind. As much as it's gotten vaguely ok to try to let the agents loose on some bits of the codebase, they simply can't even do table stakes stuff with the kinds of concurrency you see in real life.
Zig is also good at this but requires more up front design (thread-per-core, static allocation, etc.) and consistent checks to verify rules are followed.
It seems like claude code can code Rust pretty well with Opus, and I've started moving codebases away from Golang to Rust at work with Opus. Spin up an LLM and it cranks on it for a while, and as a benefit, I get easy apis to build on with other languages.
And that's the problem with Golang really, not that it's a bad language per se (all languages have footguns), but that the language interoperability story is terrible. Meanwhile Rust and Python/C/C++ go great together like peanut butter and chocolate. And I love it.
And deadlocks. "Fearless concurrency" helps a lot, but logic bugs are still possible.
Java leans heavily in the other direction: a lot of concurrency is added externally, without changing existing code, often in very declarative-flavored ways.
E.g. Future<T> serves as a foundation for a ridiculous amount of stuff, while Go forces channels for `select` whether they model your problem nicely or not, and they're very difficult (often impossible) to wrap without changing semantics.
There are very obviously lots of counter-examples for both langs (`synchronized`, rill in Go, etc), and I expect Go to become more Java-flavored in time (it already has moved this direction somewhat, and 1.27 will enable a lot more). But I think it's a fair summary of broad ecosystem habits.
1: https://tip.golang.org/doc/go1.27 (not yet released)
the source you link to contradicts your own claims.
they say:
> we’ve found bugs in every Raft implementation we’ve tested, including HashiCorp Raft, Aeron Cluster, OpenRaft, and MicroRaft
(besides Go, that's 2 in Java and 1 in Rust)
Go obviously does not stop you from writing buggy code. Neither does rust or zig or whatever. Does go make it more likely to have bugs? Or a specific class of bug? Like, the real world is about trade offs.
Yes, it's a bit of blame the user which will likely get the retort of "but I thought Go was perfect for junior engineers?"
Yes, there are footguns but none of the points therein were compelling.
But Go is also perfectly good at single threaded polling loops.
They used to blame Python a lot too - Python is slow compared to others but not so slow to matter that much, and you can build other services around it to handle certain work.
Facebook - who chose PHP - used to blame iOS/Obj-c as the reason they couldn’t build a decent Facebook native app in the early days (anyone remember Fastbook?)
I would take it with a grain of salt.
A couple more comments like this from you, and I'll be able to say, "cyanmoonx has a history of blaming the talent rather than bad tools". There being a history like that is neither an argument for nor against tools being bad. And also, don't forget that bad tools and bad talent don't rule each other out.
I've found that the LLM generated Go has few mistakes, and generally isn't too obscure. But the volume of code is so high, colleagues do a bad job of reviewing it.
I've seen a lot of very silly decisions made, like returning the wrong HTTP code, or miscategorizing a metric used for an SLO, that I just don't think is helped by the sheer volume of code one has to wade through.
Ironically, we are considering migrating some initiatives to Rust, exactly because experiments indicate it works well with LLM development.
But weirdly Opus (N=1) in Claude Code does okay on it. Enough I can reliably have it write software and feel confident it works.
This trend has been there since we started evaluating models using different languages in February 2026 and if anything, the disparity has grown in frontier models. Even Google models prefer Kotlin/C#/Rust for coming up with creative ideas (compilation success is a different story). Data at https://gertlabs.com/rankings
That being said, models love to recommend Go, and Go does have a lot going for it, especially if you are serving a public-facing website. So most of our public facing API handlers are written in Go, and we offload some of our most important binaries to Rust. There are just too many reasons not to use the languages that models think a little more effectively in.
I agree very strongly. There's no debate about things that have 1000000 permutations in other languages. e.g. The correct format can always be checked by `go fmt` with no real config options. the end.
> ...And so many languages have an opinionated formatter these days
The crux of gp's post is for Go, there is no debate as 'go fmt' is the only one that matters. Black is great, but some people prefer Ruff, leading to ...debates about which formatter the team/org should use. Go's batteries-included philosophy makes those discussions moot on so many levels beyond formatting.
The first one to choose it (whatever it happens to be) wins and that's the end of it. If it isn't the end of it you've got a talent issue.
Guess what other low-level bike-shedding argument 'go fmt' obviates? That's right - tabs vs spaces!
> If it isn't the end of it you've got a talent issue.
I know you meant this as a slur, but the implication is Go works better than other languages for those who have what you call "a talent issue"
In any case, that's a single decision the project lead takes once.
If I do the simplest possible thing that isn't a single word, by highlighting "opinionated formatter these days (e.g. Black)" and clicking search, I get the right result. I also get the right result for black formatter, and I get the right result if I yolo the entire comment as my search.
Similar to black but faster, written in Rust, by the same team who created uv.
> mean this isn't true, formatting is the most trivial part. And so many languages have an opinionated formatter these days (e.g. Black)
I don't think this is correct
Especially the uv thing. You clone some non-uv git repo that has no pyproject.toml and you don't know what to install. Maybe has requirements.txt but it's partially wrong.
> far better than C# and Java's exceptions
With exceptions you can
1. Set exception breakpoints. 2. Have real stack traces. 3. No need to write 3 lines of boiler plate every 1 line of code which interacts with outside world.
Ironically this is the case in Go's niche - devopshit, cloud webshit - where the boilerplate error handling makes even less sense since because
1. You are making a network call or OS interactions every 2 lines which can err 2. The consequences of an unhandled exception are actually quite less severe. At worst your request will 500 or the application will restart - which is a consideration you can't escape in these environments. 3. I have seen indiscriminate error handling written by substandard developers which simply concatenates the full wrapped error string to API, leaking full details. At least with exceptions, you can designate different types of exception for 404s vs ayth errors vs bad requests vs Internal errors, and have a central handler which filters out. In go theoretically you can do this, but I have never seen someone utilize errors.As over stringy error handling.
I like Go - it has nice stdlib, nice compiler and runtime, and actually dared to innovate away from the fat-ass LLVM. But error handling sticks out like a sore thumb. It's impossible to read code which does many API/network/OS interactions.
Now they are even better than me.
I do think the way software is organized for primarily agent driven repos will need to change a bit from how I preferred setting things up. (Guessing we're going to be returning to a world of microservices in the near future.)
though the letdown with Java is the wider ecosystem that makes unwarranted contraptions out of simple things.
Thus most JVM implementations had a mix of red (1:1:) and green (M:N) threads, eventually only red threads were kept in the surviving implementations.
With Project Loom now both models are officially supported and part of the specification.
Here from Oracle, as historically taken from Sun documentation for JDK 1.1.
=> Many-to-Many Model (Java on Solaris--Native Threads)
https://docs.oracle.com/cd/E19455-01/806-3461/6jck06gqk/inde...
Yeah Go is my preferred language to code with AI. Second up is type script. Followed by Java, then Python.
One of the cores behind Go is to make language simple, even if at the expense of more verbose code.
Two main examples of this is the infamous 'if err != nil' and how you handle filter/map/funcional operations.
On the other hand, being able to write 'list.filter(v => v.selected)' (or something similar) instead of:
listFiltered := []Item{}
for _, item := range list {
if item.selected {
listFiltered = append(listFiltered, item)
}
}
I'm curious to try your proposal, I just want more specifics.
I've found a lot of success pointing claude at locally downloaded docs over llms.txt URLs but not sure how to scale the pattern for a bigger project.
After all, learning a new language takes a lot of time. While basic syntax is common and quick to pick up, mastering a language's specific mental model requires a significant time investment, which is why I've used Go before but never seriously.
My interest was piqued recently when I heard about TypeScript tooling being ported to Go, and I know it is incredibly fast. However, where do the results claiming that AI agents generate superior Go code actually come from? Is it a fair, apples-to-apples comparison?
Since Go is a very small language with only 25 keywords, the way you write code is extremely standardized. Because of this, I would assume it naturally produces a lot of excellent best practices and conventions, but I'm not sure if there are actual, direct code examples proving this
I didn't say that. :)
> where do the results claiming that AI agents generate superior Go code actually come from?
Like I said - reports from users.
> Is it a fair, apples-to-apples comparison?
No - these are reports from users, not a systematic analysis.
> I didn't say that. :)
I call this the Go paradox.
I simultaneously believe we should reach for it 80% of the time to solve common collaborative problems. And being a poorer language is actually an asset in these cases.
However, in doing so, we get rusty lose our fluency in more expressive, perhaps even better languages.
How does that work? Are they generating the same project in different languages and comparing the results? What does it mean for the code to be "better"?
It's anecdata and maybe, MAYBE, a spreadsheet. Or a Google Form somewhere.
In my opinion, it has little to do with the speed of the language. The large quantity of source code to train on is quite helpful, but I think it's something else.
There are three things that I think make it well suited to LLM authorship -
1) static typing and a quick compiler - a variable can't change type after it's declared (unlike Python) makes Go more robust compared to dynamic languages. You (almost) always know what the type of a variable is. And the quick compiling with hard-stop errors means that the LLM gets a solid signal for each round.
2) It's quite opinionated, syntactically. There is generally one way that Go lang code is supposed to look. That means it's pretty easy to read as well as write. The lack of things like operator/method overloading make it an easy language to reason about.
3) the stdlib and limited dependencies. Dependency trees tend to be shallow, and because of the static linking (by default), you can generally be confident that what you wrote will run.
If I care about performance, use Rust.
If I care about iteration speed, use TypeScript.
If I want a script or numerical code, use Python.
LLMs are better with Rust because the more expressive type system provides stronger guardrails especially when writing multithreaded code. Go is almost the worst conceivable design of a programming language for LLMs: powerful but weak guardrails. Only C and C++ would be worse.
LLMs don't struggle with the low-level lifetimes like humans do. They struggle with the high-level view because of limited context windows. That's why you want a powerful type system to enforce those global constraints. Go ain't it.
Your other point is even more interesting, e. g. "before AI, Go sucked and nobody used it" - now this may be an exaggeration or simplification, but it is a great observation nonetheless, because Google suddenly tries to connect Go with the rise of AI, almost as if AI could not have risen without Go, which is indeed very strange as an argument to make by Google here. This also reminds me of Google promoting Dart/Flutter before giving up on this and preparing to send it (eventually) to the infamous Google graveyard at some point in the not-so-distant future.
It's also simply poorly informed. Go is a fantastically enjoyable language to program in. In many ways that has been a bit of its curse compared to languages like Rust (which is legitimately a not fun language to write it, and which AI tools are also very good at writing), because keeping the language simple has hobbled some edge cases.
I don't write a lot of Go as my professional life has pushed me more to Rust, but Go and Object Pascal are easily the two most enjoyable languages I've ever developed in.
Javascript takes the throne on that one.
I don't have serious metrics about if Go is better or worse than others, but LLMs seem to do fine with it.
I'm personally leaning into rust for LLM. The whole fussy compiler & errors surface at compile time seems IDEAL for LLMs for me. Hammering compile with tokens is a way better strategy than trying to deduce where stuff may fail at run time and try to catch it via tests.
Tokens are cheap, surprises at runtime are not. So a super anal compiler is what I want. I've looked at lean4 too as the logical next step but not confident I can guide an LLM competently enough for that.
My observation. LLMs find reasoning about Agda as difficult as I find reasoning about C code. I've thrown a lot of gnarly C and Ruby code at all sorts of LLMs and they have only gotten more and more impressive as frontier models have gotten stronger. With Agda, they're like "hmm, tricky" whereas for me it's an impenetrable fortress. I've asked them why they find Agda so much more difficult to write (and why they have to iterate and reiterate many many many times until they get to a destination whereas they can one-shot and two-shot C and Ruby and they tell me its the multiple competing constraints. GLM is hilarious, it flat out refuses to write Agda code but it reads it well enough. They all read it well enough. Fable is obviously great at it. And Opus 4.8/5.0 are great (if they stay on track and don't sneakily go their own way) but they're too annoying to talk to. On balance Kimi K3 is the best balance of not annoying, relatively cheap, and strong -- great model all round tbh.
So yeah, interesting I've discovered the limits of their ability coding-ability-wise. None of them are that good at designing/aesthetic judgment/architecting so thankfully they still need me in the loop.
I did a shoot out of making AI make the same simple desktop app from a SwiftUI reference for 30 different language & desktop framework combinations, and by far the best implementation came from the electron web typescript one. The least amount of LoC, the best and most complete implementation and the fastest to implement.
A bit funny, because I thought... Hmm, so LLM's find Agda natural?
My point being... It's a matter oh habit. After writing C firmware for more than a decade, I can read C easily. I might have to think of some parts and trace the code. But I can grokk it and hold the thing in my head. With Rust on the other hand, I just don't feel it as well. I am afraid writing C gave me brain damage and restricted the lens I can see the world through.
How do you bridge the mental gap?
The gap between me writing high quality rust do this steps and something being logically sound seems enormous to me
Maybe I'm misunderstanding things but I just can't articulate my ideas in casual lean4. But i can do casual rust spec
My advice is to be okay with starting small: don't go for full end-to-end correctness or anything like it. Just think of simple properties you want like 'the list returned by this endpoint should always be sorted in ascending order' or 'this operation should be idempotent' and go from there. Use your favorite LLM to help come up with example specifications from natural language, as a starting point, and try hard to fully understand those.
This kind of work does operate at the frontier of what LLMs can do, so expect to run into roadblocks (wasting tokens proving accidentally hard properties, etc).
LLMs fail to produce bug free concurrent code even for very simple cases.
Golang lacks the ability to build descent abstractions, not even mentioning the wild west of additional tools and libraries needed for non trivial micro services.
For me it is a red flag, that LLMs allow people to produce more bad Golang code faster. This is only optimization for companies which can afford enough software developers to review the excessive amounts of code needed to solve trivial problems in Golang, which are builtin in every descent programming language and/or framework.
Use LMMs and use the right programming language. This might be Golang, but most probably it is C#, Java, Python, Ruby or even PHP. (Or Rust, C, D, ...)
Go's generics are getting a fairly important improvement soon though! Generic methods, finally! It should help open up some more ergonomic patterns: https://tip.golang.org/doc/go1.27
In my experience, LLMs are excellent at finding concurency bugs.
Test it on your last concurrency bug. Point fable at the rough symptoms and ask to find where the issue is by inspection. It'll probably do just fine.
| Go is Readable / Go is Maintainable
It's true that Go, as a low-magic language, tends to be very same-y looking across projects, which is incredible for being able to reliably understand your dependencies' source code. And its tooling is world-class. I love this about Go.
But in practice I've found that, working in a monorepo with multiple teams, contributors that don't have cross-team legibility as a priority will just write SO much more code. And with business logic, often the fact that I can read the code on a line-by-line level doesn't matter if I don't understand the wider context to know how something might effect spooky action at a distance.
Pre-agents, I witnessed a fast transition from a codebase that I could mostly hold in my head to one where large swathes of it had been written and rewritten until they were unrecognizable to me. Now we have agents and, since they are still mostly not good at software engineering in-the-large, the process of knowledge debt accumulation (and ofc tech debt accumulation) in a codebase accelerates tenfold without concerted effort in the other direction. Go being easy to read does not intrinsically help with that.
A simple example is: if you highly value language popularity; Go is not most popular. If you highly value a type system that catches errors; Go's type system catches fewer errors than others. Etc. There is no weighted sum of attributes that will select Go--that's my argument.
- I had to write a moderately complex program. I didn't want to do it in C, and I didn't want to learn Rust.
- So I spent roughly about 2 hours becoming familiar with Go and playing around in Go playground. I decided that this would work.
- And then I got started on my program and I was immediately productive and that software is still running today, along with all the other stuff I've written since then.
Programmer productivity is excellent with Go. And it has a thriving ecosystem. Of course, some things could be better, but I don't really have much issue with it's error handling or types.
I thought this thread was about an ideal language for LLMs, no?
- Better concurrency story
- Native cross-compilation of static binaries (great for CLIs)
- Easier to learn, easier to teach
- Opinionated: You don't have to enforce a single style everywhere
Concurrency died out as an argument when Rust async/await got better. Sure, it has "function colors" and that matters for weird purists who care very much about typing a single "await" in their code, but don't care at all about typing "foo, err := bla(); if err != nil { return err }" all over the place. But it doesn't matter in practice, and tokio has far better concurrency tools: there ares separate channel for mpsc, oneshot, broadcast and watch scenarios, there Streams, JoinSets and a select! macro that can operate more than just channels.
The static compilation argument also died pretty early on when the Rust musl target became more mature. It's still slightly easier to get cross-compilation started with Go, but now that you we have LLMs we wouldn't care.
The learning curve argument is dead. It used to be harder to hire or train Rust programmers and that was a real pain. But LLMs don't care. The same goes for the "Go is built for software engineering" argument, which is a euphemism "Go is our way or the highway level of opinionated". LLMs do not need an opinionated language as much as humans do. If you want all code to follow an arbitrary standard, just ask your LLM to set up one. Engineering teams used to spend years bikeshedding things like brace styles and spaces vs. tabs and Go went ahead stole that opportunity from them. But this is no longer needed.
Sounds like you made a decision right there. The rest is just retro-justification, not a logical argument or comparative between options. It works for you, good.
The only logic that matters most of time is business logic of solution serving problem statement and not logic of choosing a technical stack.
"I didn't feel like choosing $LANG's competitors, so I went with $LANG" might be how languages are chosen for projects in the real world, but it's not exactly a convincing rebuttal to the argument being made.
The choice we made for other reasons is the best one for these constructed reasons that didn't exist until later.
https://en.wikipedia.org/wiki/Choice-supportive_bias
Measuring and Mitigating Post-hoc Rationalization in Reverse Chain-of-Thought Generation https://arxiv.org/abs/2602.14469
I guess if you consider enough attributes or "dimensions" then any programming languages will be the furthest in some direction, including Go.
Listing particular sets of preferences for which Go is not optimal is not sufficient unless you can show the list to be exhaustive.
This isn't an exhaustive proof as no language will every be fully Pareto optimal in practice (it's just not possible, there are too many dimensions), but I'd argue it's at least somewhat close.
One less obvious advantage: Go has greenthreading which is better than JS's async-await. Rust chose async-await to avoid the runtime overhead of greenthreading, but JS has no such reason, it's just a downside.
The tooling will let you do whatever you want, but the lack of standardization is very poor. Even now, many are switching from Eslint to Oxlint. Or Jest to Vitest. Or tsc to faster build tools. There is massive fragmentation.
I frankly yearn for a language that is always backwards compatible (JS itself might be, but ESM/CJS/etc. won’t be because the language relied on 3rd party/runtime-provided methods of importing modules for so long…), and actually has standard basic tooling from the start. Along with halfway decent security posture towards dependencies. And no need to manage a runtime. And produces smaller images.
Like if it’s just about the language, sure TS is extremely usable, async is extremely easy because it’s not a real thread, etc etc. But has plenty of quirks due to being tacked on top of JavaScript. (For example, being forced to import typescript from “.js” extension and not “.ts” with certain normal compiler settings…)
That's the main selling point, with a secondary point that it statically compiles so you don't have to do a whole Python/JS distribution thing for CLIs.
Java feels like the closest contender here, although it really sucks for CLIs due to start up times. I don't think it's the easiest to learn either, but I've never tried all that hard.
It only really makes sense to me at org-scale, though. I think you raise a very good point for individual projects, I too normally don't choose Go for that (unless I need compilation to make distribution to myself easier on corporate laptops).
Can you imagine if structural engineers rushed to post comments about how they decided to use wood for all their projects because it was simple to get up and running with, and they didn't want to have to deal with the all complexities of having to learn about metals like steel?
There’s no equivalent competitor, it’s the best if u want to just write lots of undifferentiated code.
To caveat this if u want to run about 50 agents or so in parallel, all the typescript projects burn ur disk via node modules. The rust ones take forever to compile and burn too much compute
Go can.
There’s no equivalent competitor, it’s the best if u want to just write lots of undifferentiated code.
Go is similar to popular languages like C, JS/TS, & Python. And so, easy to get started.
> highly value a type system that catches errors
Probably these folks already use even less popular ML-style languages like OCaml & Haskell; or (comparatively) obscure ones like Agda, Idris, & rocq/Coq.
Having said that: my opinion is that LLMs thrive by working in a tight loop. Unlike a human, they thrive with more and tighter constraints (and the better models are obviously far better in this regard).
I want to ditch the things that made writing code easier due to the limitations of humans, and embrace something that an LLM can leverage for better results. For me that means: an especially rich type system, (ideally pure) functional code, efficient systems-level performance and leanness. Good error messages that guide the LLM incrementally.
Go does not provide much in the way of those 3 desires, so calling it "ideal" with nothing aside from anecdotes to back that up is not compelling.
1. JS is mem-safe, single-threaded, and there are lots of training data. Easily my first choice. I'd put Python here as well, although I don't like it personally. Both should be used with "avoid external deps" in your AGENTS.md
2. Go might be a good second choice. Simple language, IMO good primitives for concurrency, well-designed std, therefore smaller potential for supply chain attacks.
3. Elixir/Erlang, little training data but rising. Safe language, safe concurrency, immutable, scalable, there are some many advantages... It has been avoided because it's different but that could change drastically in the age of LLMs.
4. Rust is probably next choice, along with C++, because while Rust is safer, the language is quite complex. C might be here too, there is a lot of training data, but every project is different and the language is very unsafe.
5. Zig, I really like the language, but it is terrible for LLMs, mainly because it's constantly changing, and the std is also under-featured and IMO weirdly designed. It's a fun language for hobby hacking, which I believe is not going anywhere, it's just not going to be something you will be payed for.
At work, we use C#. I'm shocked at how much worse it is. I've also used a bunch of Python, but I still feel the quality isn't as good as the codebase grows, but I didnt put as much effort trying to improve it.
Is Go better than CSS if you are doing web layouts? Is it better than zig if you are outputting minimal wasm deliverables? Is it better than swift if you are doing iOS specific development? Is it better than bash for OS scripting?
Think about what you are doing and choose appropriately. This was true before LLMs.
Are you having fun? Chose LISP then
The article specifically discusses how Go is well-suited for LLMs. It's not going on about general programming topics.
what would be the purpose then?
Token use didn't seem to be a criteria from my casual reading, but maybe you can illuminate me what section pointed to that, I may have been too superficial in my reading
It's hard not to, given your comments in this thread.
Which tradeoffs are you willing to accept? Zig (along with several other languages) is superior in a lot of ways for that type of job, but I still settled on Go for a particular minimal WASM (browser) project. It wasn't my first choice, but it was where I ended up because LLMs kept going out to lunch in other languages and I didn't have anywhere close to the required budget to write it by hand. I read some comments like these about Go in the past so the Go attempt was mostly a contrarian Hail Mary after so many previous failed attempts in more technically well suited languages and... it worked! Shockingly well.
It still isn't my first choice for it, but having something useful with happy users beats technical imperfection every day of the week as far as my needs go. Go really did show its worth as an LLM target for that particular workload. Whether or not that is reproducible for any other project remains to be seen, but there seems to be a growing sentiment that echos the same. There just might be something to it.
There are language implementations that would have been more minimal than that, sure, but there was no obvious way to get LLMs into alignment. I tried. Multiple times. When I switched to Go, it just worked. It may not be technical perfection, but it let me ship something I had almost given up on and it has satisfied users. The tradeoff was worthwhile for my needs. That tradeoff may not be acceptable in all cases. Hence what is best being meaningless without at least defining which tradeoffs you are willing to accept.
Two things to keep in mind here:
1. TinyGo is not Go, more Go-like or adjacent
2. 10KB still matters a lot in a lot of minimal target/usage scenarios
Exactly. Go is a language. Tinygo is an implementation, like gccgo, gc, llgo, etc. Just as gcc, clang, and msvc are not C.
> more Go-like or adjacent
It is true that recover isn't fully spec complaint at this time. That's not entirely unusual for an implementation, though. msvc is famously not 100% spec complaint with C, but Microsoft still officially considers it a C compiler, as do most who use it to compile their C code. There is usually a little grace given.
It is not like Solod that is Go-like but trying to do something quite different. Tinygo is intended to be a proper Go compiler implementation and has achieved that, aside from the recover situation.
> 10KB still matters a lot in a lot of minimal target/usage scenarios
But, of course, if the LLM cannot wrangle the language then it doesn't matter. Nobody cares how large or small your program is if you never ship it. That only matters if you are using LLMs, but since that's what we have always been talking about...
Go is behind, specifically, you have no guarantees that a given machine has Go installed, and doing stuff like gluing commands together, inspecting some files, pipe output around, or automate the boring thing in 30 seconds.
Sure Go beats bash or sh when the thing you are doing starts to become real software, but that is a problem that sits between the chair and the keyboard.
People who want to use the most appropriate tool.
> Use the ones more appropriate for what you are trying to do.
What they are trying to do is find a programming language that LLMs work well with.
> So you are using Go with LLMs for the objective and destination of token consumption for token consumption sake?
The trolling gets more intense with each comment ...
P.S. Someone else responded:
> But this isn't a user story. The user story is what you should be picking the tool for.
I don't see how this is at all relevant to my comments. I'm certainly not going to argue about what some other party should or should not be doing.
But this isn't a user story. The user story is what you should be picking the tool for.
I’ve had a great time doing LLM assisted coding in Zig, and it seems comparable to the generic Typescript/React I do at work.
I don’t doubt simplicity and good PL design pay dividends, but everyone’s favorite language can’t be the silver bullet in our new LLM world. Things just don’t add up, and I keep seeing it for Erlang, Gleam, Lisp, C, Rust, Go, TypeScript, Python, etc.
And to pick on Go a little bit, I don’t think it has any unique qualities that make it better for LLMs, where I think you could make that argument for other modern languages that offer new features leveraging their compilers and enforcing more correctness guarantees.
It would be neat to see a matrix of compile times vs. language features, showing things like:
- Bounds checks
- UAF prevention
- exhaustive enums
- test speed
But I think even among those the subtleties would make a fair comparison impossible.
Anyway I think this is all very nuanced, and anyone proclaiming language X is the language to use in 2026 lacks the experience/knowledge to consider these trade-offs and can safely be ignored.
The counter argument here is that these checks cause slower compile times and were designed to prevent common mistakes humans make.
If models get good, they may not need the same checks human written code needs. For example, frontier models already will virtually never produce a typo.
Humans need time to think, but a model’s bottleneck is in how quickly it can verify its work. Slower compile times hurt a models ability to iterate.
I don’t think we’re there yet (and we may not get there). But there is an argument to be made that languages with faster compile times may be better for LLMs in the long run than languages with strong checks but slow compilation.
"After 7 years in production, Scarf has reluctantly moved away from Haskell"
And moved to Python, pretty much for the reasons you stated
I was curious about this so I dug further, and by the author's own admission, they've only made the switch for basic CRUD logic without performance needs, not their core services: https://news.ycombinator.com/item?id=48865986.
It's also pretty unsurprising, given what we know about LLMs' style transfer abilities, that transferring parts of an existing Haskell codebase into Python would avoid a lot of the errors and pitfalls that codebases originating in Python are known for. From my experience writing lots of Python, this does not continue to hold true as you let the agents loose on your Python codebase.
~ Oreo cookie company.
(Someone who doesn't even eat cookie but heard a lot of praise of Oreo)
- There's a lot of Go code out there which the models have seen, so they know how to write it.
- Go has an exceptional standard library, so you don't need to drag in 100 dependencies to create a simple web app.
- Go compiles extremely quickly for incremental builds, which really matters when agents are building and running tests constantly.
- Go has a goldilocks blend of performance and safety. You get a good type system and excellent runtime performance without forcing the model to spend cycles fixing Rust lifetimes or Swift concurrency issues for a marginal incremental gain.
- Go is relatively stable, so the LLM's memorized knowledge is still pretty fresh (as opposed to something like SwiftUI, where the API changes rapidly).
- LLMs have clearly been trained on a lot of Rust as well
- Compile times are counterbalanced by strong compiler with excellent error messages, and "cargo check" can catch many issues without a full build.
- If you're willing to accept Go levels of performance from Rust, there's nothing preventing you from using copies and clones rather than borrows, which makes most code dead simple.
- For most major dependency types, there exists a clear "winner" in terms of community adoption, so the fact that it's not in the stdlib is not that problematic.
With that said, I have not done an extensive amount of agentic development in Rust, so maybe I just don't have the reps to compare fairly.
Nothing costs more tokens than an LLM trying to debug errors caused at runtime which doesn't map well to it's "intelligence" compared to what Rust provides
With golang, all borrow checker problems go away. This is a good trade off if your app is not cpu-bound, which most are not. If you need every last drop of performance then rust is a better choice of course.
However, I have run into a few cases of runtime null crashes in go.
You only have to compile when you actually want to test the behavior, which tends to be right on the first try more often as a result of the strict compiler.
Tencent put out this study showing that Elixir seems to reign supreme: https://autocodebench.github.io/
I'd read about this many times before I started with Go so I was particularly disappointed to learn that it was a lie. The most important task of a code formatter is to break long lines; it doesn't do it. It doesn't even have an option to do it!
My background is in data science and MLOps, where Python rules. But the focus is now less on building new AI models, and more on building the infrastructure and API calls with AI Agents. Go has a great async model, stellar performance, amazing tooling ecosystem, and far less ways of doing things than Python.
I except grow to become more and more popular, as our LLMs are now writing most of our code. Between a 50 MB portable binary in Go with 10x performance, and a 5 GB venv in Python with lack of proper parallelism, the choice is easy.
Rust or Nim are really the ideal targets for LLMs and will continue to grow. As LLMs write more code and as humans review less, it will be more important to have confidence that your code doesn't run into a weird one off runtime heisenbug issue.
That being said, the whole thing about go being "readable" is a little bit of a two-edged sword. Sure, it's straight-forward to read, but it's pretty verbose. And agents are good at producing a lot of text. The problem with reviewing go code for me is to see the forest for the trees. Subtle misunderstandings often hide in the vast amount of code that you have to read through while keeping the whole context in your head.
Anyway, it turned out that Go ironically makes it difficult to collect per-test-case coverage data. In spite of the standardized tooling it looks impossible to write a standardized collector that would run on most codebases. In hindsight using another language would have been a better choice
Joking aside, as much as Go's stdlib and tools do the heavy lifting here, Go's verbostiy and expressing simple things in lots of lines worked against me most of the time.
Maybe my problem is I'm using a language with exceptions, so trying to go against the statistical grain, with return values, is just too much.
Also, Go makes it way too easy to accidentally swallow an error. Rust doesn't have that problem.
I think partly it's because the training set contains a lot of JS, but also because complex software written in JavaScript must have impeccable architecture in order to exist at all.
It's rare to encounter a complex, functioning JavaScript application with bad architecture. I've never met any engineer smart enough to maintain a large spaghetti-code JavaScript project.
On the other hand, I've seen horrible TypeScript projects. If it wasn't for the helpful type annotations, no human being would have been able to maintain it.
These have been my and friends' observations since LLM-assisted coding started picking up steam. Go's simplicity, consistency, stdlib and tooling seem to make it very reliable for LLM generation, and it was especially true during late 2025 / earlier this year when frontier models weren't as strong; might not be as noticeable now.
As a long term C programmer and start using go from the early days, really love it's rich ecosystem and portability.
Nowadays, for any backend code, i just let agent to write using go, and for resource constraint environment, i just use rust.
And both have good C interop, and good ffi interface to hook into more higher level language such as Swift/Kotlin if one wnat to develop some mobile Apps
forbidigo is what allows me to keep ambient config out of my app, and restrict file access to a small set of paths. The coverage tool has "nocover", so you can guarantee that every realistic path is exercised at least once ("100%" code coverage, which is not a marker for testing completeness, but rather for flagging code you forgot to test). Linting is really good as well.
The only thing I haven't found is something to enforce error handling. Rust is better for error paths because you're not allowed to ignore them.
Maybe we are using different tools (or we've set it up wrong) but I'm consistently surprised at how slow Go's linting is (using golangci-lint). Takes nearly 5 minutes on our codebase after any change (which means I just don't run it locally or in-editor). It's remarkable how poor the experience is after using tools like Python's Ruff (instant) or Rust's Clippy. I'd have expected a fast, default setup that I could tune.
Event JS's Eslint, which runs in actual JS, takes 21 seconds for a full sweep (which I don't normally run, since the in-editor hints are so fast)
It's surprising, because so many of Go's dev tools are so well thought out!
errcheck, generally as manifested in golangci-lint, ensures you can't forget to do something with them. It would be odd for you to know about forbidigo but not errcheck as the former is much less widely known; is there something that errcheck doesn't do for you?
It's worth pointing out that "discard this error on purpose" is a legitimate form of error handling, so "enforce error handling" can't really constitute banning that. That's not a Go statement, that's just true in general... it is sometimes valid to just ignore the error, because there's nothing useful to do with it anyhow. I would agree the ignoring should be explicit, but it is an option.
https://github.com/kstenerud/yoloai/blob/main/docs/contribut...
Poor defaults break systems by a thousand cuts. They seem to make sense when designing the language (more convenient, less typing, etc), but then they very quickly become liabilities as project complexity increases. Go made the mistakes of mutable-by-default and silent-error-dropping, but their cyclical-import-forbidding was a good call.
> Gophers often speak of how they love that they can never tell who on their team wrote a particular piece of code—it all looks the same.
Multiple languages can have a degree of understabillity, but what matters most is context, because sometimes we need to code in a way to solve a specific problem like performance and it should be kept as is.
Another side subject I should add is about test coverage, although code is cheap, mainly because AI, guarantee that new changes to a stable code should continue to work as expected.
I worked on a few go projects with bad structure and some of them with really low test coverage (e.g. 8%), so part of the post resonates with me about we as software engineers should pursuit good architecture and other skills to allow long term maintenance.
Might be a skill issue, but I got frustrated with it on new projects constantly.
Also, I tend to ask planning questions, like "how would you implement this" and "what would the API changes be?" I'm picky about API's. Lately I've been using Deno workspaces (multi-package repos) and tell it when to make a new package or a new entrypoint. Maybe that helps?
If I just ask for features and don't look at the code, it will definitely make a mess, though. (A working mess, but it takes a while to refactor my way out.)
That's not really a typescript thing though, just an LLM thing.
Pichai wants to eliminate engineers, and DeepMind wasn't fast enough or too noble for it. Now people need to be propagandized for their obsolescence.
The biggest barrier to Go adoption seems to be Google's internal resistance to migrate C++/Java code bases to Go and refusal to admit that Go is an amazing application programming language and not really a systems programming language for bare metal OS/driver work. For example, one of the biggest barriers to Fuchsia adoption has been Google asking people to commit to Dart, I think Fuchsia would have fared a lot better as an Android successor/alternative if the official applications programming language just been Go.
(BTW Carbon isn't even a real programming language, it's still somehow stuck at 0.0.0.0 after 4 years of development which is honestly insane.)
Oh, so, little bit of self-promotion: if you like Go but is frustrated with the ergonomics of it, I would ask you to try out the programming language I developed, Oct, for LLM coding which you can kinda think of as my attempt at making Kotlin for Go's Java: It uses a codegen compiler and compiles to a plain Go binary, so it runs on everything that Go runs, and there is a lot of extra features as well: Rust style exhaustive tagged/payload enums/`match`, C#'s immutable records updated with `with`, exhaustive error handling easy parallel concurrency, xUnit.NET style unit test harness, TypeScript style compile time constraints, F# like SI unit system, Go code generation metaprogramming, etc. Would love to have some Go experts here on HN take a gander at it and provide some feedback.
I have found exactly the opposite to be true: as always, people think they can write safe concurrent code without the machine checking them and end up getting it completely wrong in lots of subtle cases. Except the problem is now much worse because you're not even writing the code, or in many cases, reading it. I prefer a language with a type system that saves me from the review burden of closely checking (and pretty much always finding issues in) concurrency invariants. And even tells me a bit more beyond that about what the code is intended to do.
That's a pile of bollocks, pardon my French. Source/proof?
And to the contrary:
I've been working on a TS codebase that calls into C++ native/wasm-compiled code for six months now. The code is mostly LLM written.
Over these last six months we had four use-after-free and two other ownership-related bugs in LLM-generated TS code.
Whereas we had zero issues of any such kind with LLM-generated Rust code that sits in another two native/wasm-compiled metacrates we use.
LLMs are not much better at ownership tracking than humans.
Especially if resource acquisition and release are far apart in code and/or somehow nested/stacked/non-straightforward.
If there is a problem with global lifetimes then the problem is certainly the person driving (or not) the LLM. Global lifetimes? FFS. Rust is hard because writing services that don’t have bugs is hard.
LLMs are not currently able to vibe a sophisticated application or service in rust. If it tells you it can do it in typescript or python, the it most likely certainly has not and you will have a wonderful time in production. Rust will burst that bubble.
And as other commenters here have said, Rust's main issue for LLMs is infectious lifetime propagation, where the borrow checker knows you violated a lifetime constraint but doesn't tell you how to actually solve it, so LLMs get error messages like:
borrowed value does not live long enough
cannot borrow `x` as mutable because it is also borrowed as immutable
lifetime may not live long enough
If the premise of the article is true, and I think that it is, that's quite the downside for AI coding with go. The premise being that reviewing now plays much more of a role than writing.
Personally, I'd rather review, say, a ruby oneliner that extracts specific row values from a csv file with filter_map, compared to 40 or so lines of go, many of which I'd have to check individually for possible mistakes.
IIUC Fuchsia uses Dart mostly for UI stuff and Go has never really tried to be competitive there? I don't see much of a reason to suppose this is a serious bottleneck to Fuchsia adoption, as opposed to the obvious reasons why it's hard to displace an existing OS with a huge install base.
Citation absolutely needed.
I think boilerplate & verbosity is an even bigger drawback with LLMs than human coding since context rot and "Lost in the Middle" phenomenon has so much effect on code quality
It seems that LLMs benefit from semantic and syntactical density
I have some very heavy criticism for Zig technically, because their whole thing about "no hidden control flow" becomes "shove all the hidden control flow into a second hard to debug runtime that runs at compile time", and manual allocation for everything is incredibly tedious and hard to keep track of in production code. I mean, C++ wasn't ALL wrong, there was a reason that templates exist in the first place, and having the entire generics model be just comptime isn't really a decision I agree with. The way I see it, Zig would probably find a niche as a language that configs C/C++ codebase at compile time instead of the C replacement they want it to be.
There are two more languages I have in the Oct repo, SDSL-V for SPIR-V shader/compute kernel authoring and Concept/Vulkan because the 20k line C Vulkan Prometheus runtime for GPU compute that we built is getting kind of unmaintainable even by AI that making up a new programming language to strangler fig refactor it is honestly the least bad option.
Each have their trade-offs, both can support native compiled application code. Some architectures are easier to review and code in golang, but others go much better with Javas richer ecosystem and better composability.
But perhaps that's also a side effect of maybe having prior opinions about go and the number of foot guns I've let off
Poor correctness guarantees, especially w.r.t concurrency. Nil pointer. Why.
LLMs are like a magnifying function. Whatever you put in you get back 10x over.
In the case of Go, in goes verbosity, Nil pointers, poor concurrency and synchronisation primitives (or poor performance of the safe ones, leading to sync.Mutex everywhere anyway). Also Go prioritises local readability over global understandability which is a poor tradeoff for LLMs with limited context windows.
So the LLM generates absolutely monstrously huge amounts of very hard to review very likely incorrect code.
No thanks.
Rust > Go.
In goes powerful, terse type system. Strong correctness guarantees not just around memory and pointers but also data races. A tendency towards using the type system to model invariants instead of relying on procedural guards and runtime assertions etc. Producing denser code is an LLM feature, it increases context window efficiency. Similarily the typesystem takes something that the LLM can spend a bunch of thinking tokens on to create a powerful global constraint. This fixes the global reasoning/context problem by pushing it back onto the typesystem.
Depending on the quality of your robot you will get different quality of code out but the ceiling is much higher. With Go better robots don't help much, even the highest quality robots output insanely verbose Go. Sort of just like with people... sort of like the language was designed as a lowest common denominator tool...
For a seasoned Rust programmer the output probably won't be hard to review, it will be easy to look at the types and either say "yeah that should probably be correct" or "no robot, do better".
You simply can't actually review the output of the slop cannons with Go, there is too much, looking at a struct tells you almost nothing about how correct the thing likely is, etc. The tests don't help either because there is going to be 10x the usual amount of those too so trying to review those for correctness is the same Sisyphean endeavour.
The thing about Go, which some have complained bitterly about and others (and TFA) have touted as a strength, is the limited expressiveness of the language (hence my remark about generics!) This is what restricts the number of abstractions in Go code, leading to more verbose but much simpler code all around. Choosing between simplicity and expressiveness is a matter of taste, but also organizational dynamics; for larger organizations which require a large amount of context shared amongst a large pool of employees, it's better for the code to be simpler and locally understandable. As TFA indicates, this has been a guiding principle for Go.
I think what is happening with AI coding is similarly related to context. Consider that while more expressive languages enable more abstractions, they can make the code more concise, but critically, this also spread the logic around. E.g. in large Java codebases you will find deep inheritance hierarchies with class and method definitions spread around a dozen different source files and JavaDoc references.
This necessitates finding and stuffing a lot more information into the context for any given task, a lot of it irrelevant and all of it more complex, because it requires making multiple hops of reasoning to figure out the logic. On the other hand with fewer abstractions, all the necessary code and logic though verbose is right there. It's much easier for a human and an agent to follow that code.
The difference is a human gets tired reading a lot of code, which is what pushes us to devise more abstractions, whereas an AI does not get tired.
I get the sense that if a context is stuffed full of highly relevant information, the agent will perform well regardless of the size of the context window. But the moment you pollute it with noisy irrelevant information, performance will drop regardless of the size of the window. (There are some papers showing this effect IIRC.) Hence simpler code, as encouraged by simpler languages like Go, are more amenable to tighter and simpler contexts, which work better for AI.
And I expect that it will turn out Python is the most productive. As it is most easy to reason about. It allows for the most elegant expression of the idea behind a program.
The first tests I have seen seem to confirm this. One recent example:
My guess is it comes down the the training data more than anything else, although I suspect functional languages will fare a little better. At least that's been my experience. There's undoubtedly a ton of python code in the training corpus and portions of it are of dubious quality. Niche functional languages likely have a smaller training corpus where a larger portion of it is better quality.
For example, Python and Typescript have the most amount of codebases and training being done on. So I feel as if that plays a part into the overall thing.
Languages which are more niche have genuinely hard times (Try arturo lang for example), so it depends on a lot of things/nuance, or well that has been my experience trying something recently.
My personal opinion is that if each language has the same amount of training. Golang comes close but the first might be Elixir. I have seen Elixir language perform really well with LLM's with magnitudes less training dataset. There have been some studies which had Elixir as the number one language for such tests iirc.
Gleam is a new addition as well and I feel as if it could be good and its another interesting option as well with more type-safety and an interesting language overall.
- BEAM makes monoliths sexy. You don't have to worry about a bunch of microservices, just focus on using proper process division for modeling your problem. - Debugging on the BEAM is first class. Drop into an interactive shell, pull up telemetry, or recon and hammer down on where your live app is slowing down if your metrics have a blindspot.
I could go on and on. I'm constantly blown away every day by the amount of time and effort and all of the sage learnings in distributed computing problems that came out of Ericsson that became the foundation of erlang + OTP + BEAM and in turn elixir + Gleam.
This sounds like your personal feelings, not quantification.
IMO the concurrency model in go is the biggest reason, I'd hesitate to use it.
Managed memory, single threaded with lots of lints and good tooling. Is IMO what can raise my confidence in code, before I even review it.
Granted golang has a really good stdlib. Which counts for a lot.
Stuff like like which compilation targets are available, or which has the most mature library for what you're doing, or maybe you're integrating with something that anchors you to a specific interface type.
Anchor your language choice to the problem you're trying to solve and the people you're trying to solve it for.
I'd read about this many times before I started with Go so I was particularly disappointed to learn that it was a lie.
Prefer standard Go libraries and tools.
80% of the time I can get by without external dependencies (outside of Go's X repository)
Not as fun to write as Python and Nim, but I don't have to write it.
I still think Go is a very excellent choice but I have switched to, of all things, AssemblyScript within a Rust host. I've been very happy with it - surprisingly so. Compile time is a major drawback of course.
https://news.ycombinator.com/item?id=47222270
203 points | 5 months ago | 304 comments
I think the most important thing is how big the standard library is. Pulling in 3rd party dependencies is where I begin to lose a lot of faith with LLM authored code.
As a lead I'd love to use rust, I will put in the time on my own, my team won't or can't. They treat this like any other job they signed up to deliver value with what they know. For hiring not everyone has the talent pool and fund access to get the goat-ed engineers that congregate to tech hubs for maximizing their income. Then if you get through that cherry on top is LLM's are only as smart as you guide it to be. There is probably a staggering amount of ways to write 1 approach to business logic, you may not know the ideal pattern so you'll commit to a worse one on the company dollar.
I'm moving my team's projects slowly to go because, its easy to go from novice to advanced in terms of code writing,legibility and patterns. We also don't have deep ecosystem requirements to ts/python in most of our work. It is verbose but I don't mind that on token spend if it gets done with with validation/error handling which it obnoxiously enforces. It runs cheap, ecosystem is good for platform eng, standard library does a ton out of box.
If a language is simple, it' easier to generate good code.
Even as a systems lang, the error syntax is the worst part of Go. Can they at least put the ?/! syntax like in Rust instead of this "if err != nil" spam every other loc?
I never seen k8s cluster that doesn't have some go process that segfaults once in a while because someone forgot to check `err`.
Only good thing got going for it is its vulnerability scanner. Which will be working overtime with all that "AI-assisted software engineering"
> Go solves this through unyielding consistency.
What? Why is the word "unyielding" used here? What was the point of generating this AI article on the google blog post?
Google.
Now one can say that a programming language and its design or usefulness is - or should be - decoupled from the company developing is. I am not opposed to this, in theory, but Google goes way too much on my nerves these days. And I am hardly the only one here.
I am not saying this is a rationale used by many other people either, mind you, but Rust has been taking strides (not that I am a huge fan of it either but for different reasons) and it seems to me as if Rust has finally now more momentum than Go, which I find interesting. Again, this may be a correlation rather than any causation, but I can not help but notice it.
If you're going to have an agent write most of your code readability is very important.
Go's historic maintainability strong suit has been its simplicity and consistency. The syntax is, relatively speaking, lightweight, the language invites complexity through composition, and information density for any unit of code is typically quite low (which isn't necessarily a bad thing).
In my opinion, though, these are all drawbacks, and Rust addresses all of them. It's syntactically and semantically much heavier, leading to its oft-maligned steep learning curve. It has, uniquely among the major languages, I think, a syntax for expressing variable lifetimes (with its own unintuitive semantics). It stuffs lots of abstraction into a hodgepodge of terse semantics and punctuation.
It sucks to read, until you get really used to it. Then it tends to read really quickly, and, at least for me, it's easier to reason about a conceptually-broad piece of logic if I don't have to jump between different locations in a file, a module, or a package to do it.
With Go, I find it more difficult to get into a flow state, and easier for my eyes to glaze over when looking over large diffs.
It's not lost on me that these are purely subjective arguments, though. My preference remains with Rust, and that goes back to before I used LLMs.
I'm also aware that Go is very prescriptive about how you write it; it's explicitly opinionated, and Rust doesn't have that. It means that most Go code bases will look more alike. I consider this an anti-feature; I believe code should be able to conform to the problem space or product and a good team will find the best way to do that.
I think it's the use of pointers and "if err != nil {}" error handling spam. It reads as a highly compromised imitation of Python and C rather than a solid execution of some other idea.
Rust is not the most beautiful language out there but it doesn't trigger any such reaction for me. The ? operator and "match", which I use constantly, more than compensate for some of the sigil noise which I barely need to look at much less write most of the time. So Rust wins on that comparison for me.
The "func name() -> retval" syntax also grew on me. I like the fact that Python type annotations copied that approach, and C-style declarations look ugly to me now. Same with C-style /* */ comments.
Rust is definitely jarring to look at, in the same way that decoding some strange C declaration can be, in ye olde days when you had to float all this context in your mind while doing work. But with modern tooling who cares: "explain this lifetime to me"
You can pretty clearly see the limitations if you read, for example, the type of code the Protobuf compiler generates when trying to compile Protobuf/gRPC enums or structs into the way-more-limited Golang type system (this is despite the two being designed to work together). And it could really do with algerbraic data types and other modern programming language features.
Also the type system does have a couple weird behaviors that seem straight out of JavaScript. Like the difference between struct and interface nil for example:
```
var buf *bytes.Buffer = nil
var out io.Writer = buf // now out is nil
if out != nil {
// This block will execute because out is not nil
out.Write([]byte("crash")) // This line will crash because out is nil
```
Many things about the language almost seem to be designed to simplify the implementation of the compiler rather than to benefit the developer experience.
E.g. make a table that's 3x3 is easier to read (Go), but the equivalent line in Rust would also include material, angles, height, etc. because the type system encodes much more information.
Though I always found Go to be significantly harder to read than Rust. Sure Rust has some crazy syntax at the edges, but Go makes it very hard to know where imports come from (and thus what they do), and the imperative style + lack of clarity about mutability makes code much harder to reason about.
I my self and teams members are slowly reading less code and requiring agents to prove things work the way we want in other ways.
The default sentiment is humans should read code it's more progressive and a leap of faith to start giving that up.
Obviously, I get why you feel humans still reading code is important, but if you look at the progress of AI for the past couple of years, that gap is closing. The trendlines speak of a future where it becomes less and less important.
This was exactly what happened with writing code. Now most people don't write code.
I use LLM daily to write code for and "with" me, I also write code without LLM. Most people I come across mix it up. A few do it all by hand, and equally few all by LLM I would say. Is that just in my corner of the world?
My entire company for example does not write a line of code. We manage agents and that's it. Many, many, many companies and people are already doing this.
Go is simple, no “magic” marcos or meta programming even with just a little programming in any language it’s not hard to understand what the go code is doing.
I have no doubt that you can get a LLM to write working bug free code in any language but that is not the topic of the article or my comment.
No way to know until someone does the science on this. Until then it's just people saying that more static checking is better. But I do think, anecdotally, python is horrible for LLMs.
Of course it can’t just be simplicity, it has to be “opinionated” simplicity. Rolls eyes.
Let's assume that you need to write a program with a given set of requirements, and that you have a magic wand that can instantiate a high quality implementation of the program in any programming language instantaneously and for free. My hot take is that you would not want to choose Go, and you would likely want to choose Rust.
The Go implementation will have higher memory and CPU consumption due to garbage collection, while still being subject to memory bugs. The Rust implementation would be as efficient as possible on the given hardware with minimum memory/CPU, and it would be immune to memory bugs.
In my view, the biggest challenge with Rust, and where Go wins, is the relative difficulty of writing in Rust as the language is significantly more complex. With LLMs this is becoming a non-issue, and we are getting ever closer to having this magic wand (I'd argue that for smaller programs the wand already exists today). The article advocates that Go has excellent readability. I agree that Go has trivial syntax, but given that it's so verbose, I actually find it easier to read Rust code. Its higher expressivity allows you to see the higher level intention of a piece of code more easily.
Many of the other benefits the article mentions for Go are equally applicable to Rust: compiler error messages are super detailed and a great help to coding agents, auto-formatting, a great language server, and a package ecosystem.
I guess from a pure language POV, one might argue that rust leaves humans with more opportunity to add abstractions that are too clever and too hard to wrap your head around. That doesn't feel like a strong argument, though.
Then there is of course the ecosystem, where go maybe has better libraries for some stuff (while rust may have better ones for other things).