mq

Specs, Intent & Agentic coding

[mq @mio___q]

@harrysolovay @samgoodwin89 So if it was effect atom / signals based you would change your mind?

[mq @mio___q]

@stevekrouse It needs to be less precise than code to be of any value.
It should capture intent, and be iterative as you say

[mq @mio___q]

@dillon_mulroy Zig?

[mq @mio___q]

@Jaidcel @ThePrimeagen Omg. This is beyond stupid.
FP goes incredibly well with AI

[mq @mio___q]

@kentcdodds I dont know, which challenger offers sharper DX?
They're pretty much identical at this point in terms of usage

[mq @mio___q]

@theo Would love to hear his take on React 😂

[mq @mio___q]

@RhysSullivan Ah you must be using formal verification

[mq @mio___q]

@jessoclarence @zeeg What does the orchestrator do here exactly? Would be keen to hear in more detail

[mq @mio___q]

@MichaelArnaldi Yeah, I think the whole premise is flawed.

I dont want random activation and have to write "dont use skill X" every time I prompt. This happens every time I decide to try even 1 skill, and I have basically 0 installed.

[mq @mio___q]

@IntuitMachine I think it can also be explained by incentives, people trying to protect their investment

Whether that be their AI startup, job, skills etc.

And they perceive anthropic as a threat to that.

[mq @mio___q]

@RhysSullivan Gpt-5.5 xhigh is no longer allowed to touch my code. It acts as a dumb middleman who simply hands the task & context to GPT Pro, and applies the patch it gets back verbatim

Apparently even this is too challenging for it to handle reliably

[mq @mio___q]

@dillon_mulroy Same reason I dont use skills 😀

[mq @mio___q]

@mattpocockuk It starts at ~0K with GPT 5.x

[mq @mio___q]

@MichaelArnaldi @OpenAI Yes. Its pure evil. I feel scammed

[mq @mio___q]

@thdxr @JakeKAllDay So just as before, until they nerfed it?

[mq @mio___q]

@MichaelArnaldi Been for weeks like that for me 🙁

The only way I can make any kind of progress is:

Implement
-> Revert, Repent & Reflect
-> Reimplement

[mq @mio___q]

@LobozDev @MichaelArnaldi @thesherlocker Nah, both are the same. OpenCode is slower but at least you dont get routed to "dumb mode" as often.

[mq @mio___q]

@FUCORY @benswerd @e2b @mlejva Seconded! I would have switched to it already if it wasnt for that

[mq @mio___q]

@mitsuhiko We should have a social credit score for this

[mq @mio___q]

@jamonholmgren I noticed it when I do that it always seems to route me to fast/dumb mode with:

"You're right <wall of text>..."

After that never again... Id rather punch a wall than that