I used to really like Warp, back when it was just a terminal. Then as they went more and more in on AI the actual terminal features became more and more buggy and I ended up having to go back to WezTerm.
Why is every company seemingly trying to re-implement agents themselves rather than making their product work with the existing ones?
Yeah I became a fan of Warp when they offered a terminal that treated the input as a normal text input field so I could use my normal keyboard text editing shortcuts.
But I understand that's not a viable business model. I'm back on iTerm2 now cause Warp isn't approved for use at my job (and never would be with all their AI pushing), and I've managed to pound iTerm 2 into something usable enough but I wish every terminal would copy Warp's text input style.
I never used Warp, is the editing similar to PowerShell ISE's editing by chance? Powershell ISE is like a mix of a script editor and a PowerShell UI so you can script out some things or use the terminal directly.
The killer feature that got me into using it was the shell integration. Your prompt, input and command output are entirely separate entities rather than one single stream of text. It's like using a Jupyter notebook for the terminal.
That was also the thing that destroyed it for me, since navigating between history entries started getting more and more unreliable and sometimes clicking on the input box would randomly resize the entire window or move it between screens. I don't entirely know how they managed that one.
does anyone know if the above experience and fluid vim-style behaviour can be replicated on something like ghostty? this was the killer feature of Warp imo and I saw no other benefits, but somehow not other terminal has been able to (or tried to) catch up
I also felt this way and after staring at the 900MB app package in disbelief started tearing out features using Claude with minimal supervision over the weekend.
It was also a bit of an experiment with a larger Rust codebase, which is outside my usual areas. I did pull out powershell since this was just for personal use on macOS. Been using it today and enjoying it in place of Ghostty, but merely an experiment and learning project:
The writing was on the wall from the very start of Warp. I can't remember precisely what happened, but years ago when it was first announced there was something that required internet, or telemetry, or an account to use Warp. I noped out right there before I even tried further. It was always going to be bloatware.
100% agree. I've ended up building my own [1][2] mostly because I got annoyed the agent cli's are building remote control into the agents, instead of into the terminal. I really want the agent cli's to focus on being the best autonomous agents, then let me choose a great terminal that does all the extra.
Another great option (not mine) is Orca [3] and it's a lot more mature. Really nice to use.
I'm glad it works for you, but I've got enough electron apps running on my computer already. I have to run a bunch for work and fully half my cpu is just electron at idle
Another reason why individual developers do not pay for their tools. This is one of the only professions that does this and they always pass the cost to the employer, who will pay for it until they cut costs.
So the open source code on GitHub is the library that agents need to train on, to do all the economically useful work for 1/100th of the cost or close to free.
That is where it is going and the raised money will be going to the agents.
when i open a new tab in warp now it defaults to their agent and not my actual terminal. i'm going to have to switch back to something else, do you like wezterm?
I don't dislike it. The keybindings are a bit arcane and having to write the config in Lua rather than having a preferences panel is annoying (though spectacularly powerful), but I've settled in to it and being able to use ctrl-shift-p to control the entire terminal is nice.
My absolute favorite thing about Warp is the function that suggest a new command intelligently when the one I just ran failed or changes to the code when execution failed. Haven't been able to replicate that as smoothly in Ghostty which I really want to enjoy but honestly haven't really felt the same delight I got from Warp.
> Cloud agent handoff
> Start work in the CLI, then hand it off to the cloud when you want to close your laptop. All cloud agents are tracked centrally and can be monitored and steered via the web so you can continue work on the go.
They hype up not needing a third party binary to remote and run agents against SSH'ed systems, but how does this "cloud agent" work then? Are they shelling you through their own back-end? Because that sounds bad and probably wrong? Maybe its a different use case.
The other day in warp I literally couldn't run `ls` because it kept interpreting it as a AI command. I still love warp, but some of the stuff is crazy.
So this is a competitor to something like OpenCode or Pi? I suspect for most developers, they are using Claude Code or Codex CLI mainly just so they don't have to pay-per-token and can use the subscription products - which I assume can't be done with Warp Agent?
I suspect most developers don’t care too much about cost because they’re just using what their company is just paying for whatever enterprise subscription/billing.
The codex and many other subscriptions can be used with other harnesses like OpenCode and pi. It's officially supported most of the time. Using my Codex and MiniMax sub just fine in pi.
I do like integrating the shell with the agent. That seems like something I do occasionally need. But the occasion when I need it is getting rarer and rarer so I suspect that this is building a feature for 6 months ago.
> think of our CLI agent as a built-in mux’er across agent sessions. This allows more natural interactions, like being able to switch directories in agent sessions, have the agent drive full-screen terminal commands (e.g. sqlite and mysql), and even run across ssh sessions with no remote binary install.
Looks like they let you use your LLM of choice on anything your terminal can SSH into using their TUI without having to install Claude Code or their agent on a remote system. Also, you are able to have multiple shell instances as you would on tmux in one session, all with the LLM fully aware of each one (not sure if its a model per session, or multiple, or what).
No guy, this is the one, this is the coding agent that everyone is going to use now, instead of the one that is in the editor or part of their existing ai providers offering.
The irrational exuberance of these guys. It's interesting to see what happened, somebody somewhere got the very very bad idea (if the goal is to make money) of "we will build, like the BEST terminal."
Which, given open-source, is just going to be a dead-end, even before AI. Because software likes to be modifiable, and this sort of thing is literally at the center of people who like to modify software.
But, hey, if something good comes of it, fantastic. I mostly the think of Docker, an unqualified success in software and probably a huge failure in business.
And the latter actually makes this a double-good; a "successful" docker would have somehow monopolized some very important part of the process, and docker is too good to be gatekept like that.
Why is every company seemingly trying to re-implement agents themselves rather than making their product work with the existing ones?
But I understand that's not a viable business model. I'm back on iTerm2 now cause Warp isn't approved for use at my job (and never would be with all their AI pushing), and I've managed to pound iTerm 2 into something usable enough but I wish every terminal would copy Warp's text input style.
That was also the thing that destroyed it for me, since navigating between history entries started getting more and more unreliable and sometimes clicking on the input box would randomly resize the entire window or move it between screens. I don't entirely know how they managed that one.
It was also a bit of an experiment with a larger Rust codebase, which is outside my usual areas. I did pull out powershell since this was just for personal use on macOS. Been using it today and enjoying it in place of Ghostty, but merely an experiment and learning project:
https://github.com/dxmachina/warp-nine
Another great option (not mine) is Orca [3] and it's a lot more mature. Really nice to use.
1. https://github.com/markwylde/terminay
2. https://terminay.com
3. https://www.onorca.dev
You can't raise a $50 million series b on making a nice terminal. we already have lots of nice terminals for free
So the open source code on GitHub is the library that agents need to train on, to do all the economically useful work for 1/100th of the cost or close to free.
That is where it is going and the raised money will be going to the agents.
Why does any company do anything? Money, of course.
They hype up not needing a third party binary to remote and run agents against SSH'ed systems, but how does this "cloud agent" work then? Are they shelling you through their own back-end? Because that sounds bad and probably wrong? Maybe its a different use case.
> please implement the fix
$ please: command not found
I've read the product description and I don't see anything that's truly unique
Looks like they let you use your LLM of choice on anything your terminal can SSH into using their TUI without having to install Claude Code or their agent on a remote system. Also, you are able to have multiple shell instances as you would on tmux in one session, all with the LLM fully aware of each one (not sure if its a model per session, or multiple, or what).
I've disabled it all because it gets in the way.
Which, given open-source, is just going to be a dead-end, even before AI. Because software likes to be modifiable, and this sort of thing is literally at the center of people who like to modify software.
But, hey, if something good comes of it, fantastic. I mostly the think of Docker, an unqualified success in software and probably a huge failure in business.
And the latter actually makes this a double-good; a "successful" docker would have somehow monopolized some very important part of the process, and docker is too good to be gatekept like that.