JanW

Doing AI things for fun and for living.
Blog
Refurb laptop watcher
@janwunschcz Czech-only X

[JanW @janwunschcom]

@leeschmidt123 @blitzbohne @ericintheloft @vercel Bro are you paid for this? :D

[JanW @janwunschcom]

@leeschmidt123 @blitzbohne @ericintheloft @vercel Only reason every indie dev needs lol

[JanW @janwunschcom]

@thecmdrunner Looks like a hotel room lol

[JanW @janwunschcom]

@dhh I recently upgraded to starlink as I don't have any other good options where I live... and here I am with my 100Mbit starlink thinking "finally I have a decent internet" 😅

[JanW @janwunschcom]

@jankeesvw I listen to ambient music for programming, like this guy's channel is superduper https://t.co/kGhUsnUPg9 . Or at Spotify this playlist is great all the way through. https://t.co/ChXicxiXlo it makes me feel like being the hero of some sci-fi cyberpunkish movie, just the vibe I lov

[JanW @janwunschcom]

@levelsio @dillon_mulroy @X Will you then stop blocking non ai accounts from not being able reply to you? :D

[JanW @janwunschcom]

@jmilinovich Did you have ports opened in?

[JanW @janwunschcom]

@_ongedierte_ @theo 😅 huh I would love to hear more here

[JanW @janwunschcom]

so i've tried t3 code and honestly i'm not sure what's the deal. i'm not switching my tui herdr - jcode setup to t3 code.

1) i am still much more comfy in terminal and keyboard - matter of habit i admit.

2) only thing it does natively better than my herdr - jcode setup are native mobile app (there are good external solutions for herdr though) and the webhooks (can be solved easily if the need arise).

3) it's way more memory hungry because it uses the memory hungry clients. i work with jcode and i can run 20+ agents and it would use around 1 GB of RAM. using t3 (or claude code directly, essentially same) i get that usage at around 3-5 agents. this is super critical for me.

i tried it very briefly and i may try it further, perhaps i'm missing something obvious but so far no extra value i see in GUI over TUI, despite @theo's being pretty vocal about how terminals are holding us back. prove me wrong

[Theo - t3.gg @theo]

It is insane to me that there are still devs using terminals to prompt their agents

[JanW @janwunschcom]

im not saying all 20 agents are running heavy builds or nodes or whatever. the problem starts when even idle agent uses >150MB of memory instead of 30.
ofc harness is not my primary solution but if you give me 2 solutions and one is 5x more efficient, chances are im picking that. i like efficiency every day even with 30 GB of free memory.

[JanW @janwunschcom]

@Igloczek @theo Oh and as a bonus I personally love - I don't need to touch a mouse to controll all this :D

[JanW @janwunschcom]

I'm not sure what you mean by few agents or where does the assumption come from. This morning I've had in my herdr 27 separate agent panes running across 5 different projects + some of the agents were implementation coordinators and each swarming 2-5 of their own agent workers (which I don't need to follow closely but if needed I can see them in separate pane on demand too). I deal with them fine thanks to the spec/ticket workflow I use and obviously herdr pane logic and not sure how exactly would T3 improve that. Enlighten me if I'm wrong but the agent management T3 vs herdr is very similar.

[JanW @janwunschcom]

@ABorealls @theo Curious about that too :D sure I'm just nobody with 19 followers but the work I do with jcode feeds my family pretty well

[JanW @janwunschcom]

@JeffScottDev Why?

To be clear I tried that but did not see nothing herdr wouldn't do

[JanW @janwunschcom]

@BurmLabs You should refund my 5 s of my life reading this reply +10 s for writing this

[JanW @janwunschcom]

@Igloczek @theo What does that mean? Explain pls

[JanW @janwunschcom]

@digitalshane_ Wtf why x thinks your post is Tagalog 😅

[JanW @janwunschcom]

@Igloczek @theo herdr works at multiple devices just fine too

[JanW @janwunschcom]

@dimaip @theo im just in process of moving my coding env to minipc sitting in rack with enough memory and this won't be an issue anymore.
still though im a bit of optimazation freak and even when solved i'll be probably judging harness by its memory usage :D

[JanW @janwunschcom]

@theo ofc the dev loop eats most of the memory in the end but on my 16 GB RAM machine it makes huge difference wether harness uses 1 or 4 GB RAM for the same amount of running agents. virtually it's the difference between OOM killer and running machine