Erik Kovari

Co-Founder, CTO @lissenlive
Lissen = streaming + artist-to-fan 🔥

[Erik Kovari @erikkovari]

I am incredibly boat-pilled, love the product. Been using it for all kinds of stuff. But the limits are a huge issue - we are barely using the time way pay for, and only pay to have the higher limits. Those limits mean I can't use it for CI. @AniC_dev do you have any plans to allow paying for higher limits rather than time?

[Erik Kovari @erikkovari]

@MichaelH_rley @ParthJadhav8 @jullerino @t3dotcodes What are straight-to-background threads?

[Erik Kovari @erikkovari]

@dorian__mazur @egorvodopyanov Every commit? That sounds very slow and/or expensive. Curious how you deal with that?

[Erik Kovari @erikkovari]

OAI was swimming in compute, their models were really efficient, but less smart. Now astra is super hungry, they don’t have the compute.
Ant is the opposite - they made the best models, they were hungry, now they figured out how to make them efficient and now they have enough compute.

But I agree, it’s working out in Ant’s favor - I doubt that any of this is intentional strategy though on either side. On OAI at least, not allowing new 20x signups is surely not something they want to do

[Erik Kovari @erikkovari]

@FlyaKiet AGI should be versioned, its software. I’d estimate we’re somewhere around AGI 0.0.2-beta

[Erik Kovari @erikkovari]

@tibo_maker that 'take over' feature on the Hero is crazy cool, never seen that before!

[Erik Kovari @erikkovari]

@dschewchenko lol, just saw the same error, came to X to see if they are deploying Astra and this is the first post I see

[Erik Kovari @erikkovari]

@ThePrimeagen @zeeg And then spend 20 minutes trying to figure out how to exit that session

[Erik Kovari @erikkovari]

@theo or @jullerino any idea why the 'reset to this message' button is greyed out for me in @t3dotcodes? it seems to be like that in every conversation? this one is Fable

[Erik Kovari @erikkovari]

@BenLissen @t3dotcodes @theo as it is usual with my lovely co-founder, I have no idea what you mean :D afaik you can generate images and code on both cli and gui, wether you're using codex or t3code

[Erik Kovari @erikkovari]

I just tried @t3dotcodes finally, and I am converted to @theo's church immediately. I have been working like this:
- Codex app for gpt
- Superset for claude
that is because the UX and remote is great in codex
but this is incredibly annoying and requires me to juggle a lot especially when using remote from my phone (seamless in codex, sucks ass in claude app).

I love superset, but with the direction these model providers are going, it's making less and less sense to use their cli, and I still can't do remote with superset because there's no app.

having one app that works with the same UX as codex but across different models AND REMOTE is awesome.

couple of other things I really appreciate:
- perf. I don't know why, but typing on t3code just feels a lot faster, it's the first thing I noticed - is there anything special happening there?
- perf2. recently codex has been really slow, and t3code is a breeze in comparison. let's see if it starts slowing down one I have hundreds of threads.
- t3code saves your 'draft' thread, amazing. in codex, if you start typing a new chat but don't submit, and you click off, it's gone - super fucking annoying

[Erik Kovari @erikkovari]

@FlyaKiet @t3dotcodes @theo I know, you liked my comment where i was asking for an invite, the invite never came 😢 i'd still love to try it though!

[Erik Kovari @erikkovari]

@oorestisime @thdxr @CodeAkram When GitHub is down, blacksmith is down - they are just providing the runners, gh still orchestrates the actions

[Erik Kovari @erikkovari]

I generally agree with this but I challenge the definiton of ‘maintainable’ and ‘deep understanding’. Before our machines could write the code, these words had a different meaning in my opinion. To make something maintainable meant clean, documented code that a human can read. But now the machine can understand it faster even if it’s ugly. Similarly, we had to have a deep understanding of our codebase otherwise we have no idea how to fix something that is broken; but again, the machine can do that now, and it can give us complete understanding much quicker.

So if we accept that we can build good enough systems (which, i am not claiming we can easily) to work with the agent safely such that we in fact don’t need ‘clean’ and ‘maintainable’ in the old-sense of those words, and the code produces the outcome we intended and it works, then why do we need to read all the code?

[Erik Kovari @erikkovari]

@jmeistrich Can’t believe you did it, thank you 🙏 will definitely try them

[Erik Kovari @erikkovari]

@FlyaKiet @dodeja @superset_sh @openclaw Me too 🙏

[Erik Kovari @erikkovari]

@chuckmarvin @danshipper @every Make agents.md symlink to claude.md both in local and global and job done?

[Erik Kovari @erikkovari]

@JustinBleuel I will feel Terrable if it doesn't release today!

[Erik Kovari @erikkovari]

Been thinking about what Sonnet 5 is for, I think I get it:

Fable is awesome, and genuinely much smarter at finding bugs, implementing, using tools. But it also does a LOT by itself, in each turn and doesn't talk to you cleanly about it. It just moves on to the next thing without saying anything - successfully - but it's way more of a black box.

So Fable is much better for one-shotting huge work, but if you need the back and forth, continuous hand-holded way of coding for a real reason, Sonnet 5 is better at that than Opus 4.8

[Erik Kovari @erikkovari]

@gabimoncha @t3dotcodes what's the difference?