Dusan

manual incident response is cooked. I gave Claude the /traceway skill and it fixed a production panic in under 2 minutes. Open source. MIT. No vendor lock-in. 3am wake-ups are now optional. Demo video + repo in replies 👇

I don’t know what implementation of promises you’re referring to. Give me a specific language/implementation because I would have said that sentance in reverse ‘you can layer promises on channels’. The way I think about async await is as syntax sugar for a queue of tasks that execute, where a resolution callback (success or failure) is added to that queue. This would be trivial to implement with a channel, I don’t see how you would implement channels on promises. But we might just be looking at promises/async/await in different implementations. What language/framework are you referring to specifically?

Channels are explicitly a form of CSP. But the interesting thing is that go also has atomics and shared memory. I spend way more time with atomics (locks) than channels. But I'm specifically talking about a language primitive, not the abstraction and patterns built above. You can layer channels on promises.

After many years with go, I don't think channels are a big win. Sorry.

Channels are more of a communication pattern between code that might be executing on different cores (in my mind) than anything else. I view them as a communication mechanism. I’d say that actor systems are a solid alternative but def not promises. I don’t see how an async/await w a promise could lead to similar results, but maybe it’s lang specific, which implementation did you have in mind specifically?

Promises are a better abstraction but what people conflate is that go has two features. It's fundamentally a fiber that yields on I/O and then the concurrency primitive is channels. So just other languages that do async/await use it for I/O. The innovation of go is to treat I/O and concurrency as two distinct things, since I/O is the core reason for async getting scattered everywhere in a program. You shouldn't have to treat I/O as concurrency, you should only treat concurrency as concurrency.

I'm happy with it 30-60% of the times, but I also try to give it enough info about the code itself that anything less than that would make it borderline unusable. Like I will literally tell it what functions/interfaces I want it to add and how I want it to connect things. Like a smarter auto complete. I think it's too dependent on what you're working on as well, for example it's really good at modern web boilerplate but really bad at complex coding or less popular technologies. It also sucks with custom built frameworks/libs. I adjust the amount of code I expect it to generate based on the task as well, example: I give it less details for a react/svelte component in fairly standard project and way more info for a legacy ember project to get similar results. Idk, just my experience.

How frequently do you estimate a model produces a patch you're honestly happy with, or at least have confidence in no longer term issues? Real world scenarios, not your hobby projects.

@ImLunaHey you just saved me $400/mo on RDS instances I forgot to delete

reminder, cancel all of the subscriptions you're not using.

reminder, shut down RDS instances you're not using

@paulg I think people are reading more than ever. I can tell the programming section has basically gone extinct at my local B&N, but everything else is thriving. What leads you to believe people are going to stop reading books?

Something I told 14 yo: People are going to stop reading books. I wish this wasn't so, but I fear it is. The silver lining in this cloud is that if you're one of the few people who still read, you'll have a huge advantage over everyone else.

now I feel bad for Zig, maybe I'll end up using it after all

@ImLunaHey literally wild how fast you're moving

really hoping i can get my js engine lumen to near node/bun speeds before the end of the month.

@rob_mcrobberson based

if you believe all that j-space nonsense u have ai psychosis the swap experiments are real interp. fine. good work even. but the “claude secretly built a conscious back office nobody designed” narrative is absolute nonsense touch grass, read the methods section

sigh, was gonna use zig before the blog wars, guess we’re going back to the drawing board

@ibuildthecloud Maybe they find Doofenshmirtz irresistable

@dduxAdventure The bots just like me better I guess.

These bots come and go in waves.

@16vchq it runs out of cash. every. single. time.

Finish this: ‘The startup dies when _____.’

@ImLunaHey make sure you announce a blog post about lumen and drag it out for months

still not as fast as node or bun but were getting there.

@MFrancis107 the duality of men

I'm pretty sure Fable is more expensive than a senior software engineer in most of the US. It's not worth it.

@karanbhilhatiya if it goes up 2x I'd probably drop it lol... I'm on the $200 monthly plan

if claude and chatgpt will cost 10x more tomorrow, would you still use

@dark_coderz what a dummy, giving it input, everyone knows that loops are how you do dev this week

met a dude, he still copy-pasting the code from chatgpt

@BratDotAI programming

What’s the one thing Claude consistently does better than Codex?

@AlexanderTw33ts yes, make sure the todo app is secure before it's too late and you're stuck in the permanent underclass

POV: anthropic team allowing the permanent underclass to have 5 more days of sota intelligence

@sflorimm y'all are sniffing some medical grade hopium. if your todo app goes down nobody cares, but we don't need any vibe coded bs in any infra critical piece of software, for example: medical devices, airplanes, cars...

why do some coders still refuse to use AI?