Tried @orca_build and @t3dotcodes for a week each.

Tried @orca_build and @t3dotcodes for a week each.

@t3dotcodes wins by a mile for convenience. @orca_build wins on integrations.

My main use case is running all my coding agents on a Mac Mini that I keep somewhere else and connect to remotely. So a lot of this comes down to how these tools fit that setup and my normal workflow.

Right off the bat, T3 Code has a much better and more responsive UI. The experience with T3 Connect is seamless compared to Orca Relay.

The problem with Orca Relay is that it acts as a pass-through layer to the terminal on that machine. You don’t get native text inputs. Paste is wonky, typing is wonky, and sometimes the text just doesn’t get sent through.

What I really like about T3 Code is that I haven’t had any of those input issues because the UI layer handles them. It captures all of your text and voice inputs and then sends them to the other machine. That is insane and really good.

With Orca, that terminal experience also means inheriting the terminal flicker issues from the CLIs you’re running. Cursor’s Agent CLI, for example, has not been a great experience for me through Orca. If the thread gets too long, the terminal flicker becomes unmanageable. And on mobile, the TUI doesn’t render properly, so you have to wait for it to stabilize before you can do anything. That’s annoying on its own.

I haven’t had those issues with T3 Code.

The next thing I like about T3 Code is that I can work the way I normally do. I usually start by going to the repository, opening the main branch, and making sure it’s in sync. Then I create a new branch and start working. I don’t usually bother with worktrees for that workflow. To me, it doesn’t really matter because worktrees take up too much storage when you have too many of them, and they become a pain to manage.

However, I like that if you do want a worktree, you can set one up in T3 Code and it works seamlessly. It’s there when I want it, and I can still use my usual branch workflow.

What I love about Orca, though, is how well it supports having multiple agents working on the same branch at the same time. I take @steipete’s advice about blast radius super seriously, and I split my work in large codebases accordingly. I always consider the blast radius when deciding what each agent should work on. Orca supports that way of working much better for me.

Can I do that in T3 Code too? Yes, you can. It’s just that the UI feels more suited to multiple agents running in separate worktrees. Having multiple agents working on the same branch feels more natural in Orca.

Then there are the integrations, which are a big part of why I still like Orca. I have one Claude Max subscription, one ChatGPT Pro subscription, and a Cursor Ultra subscription. I want to be able to use those subscriptions in my workflow, so the choice of harness matters to me.

My problem with T3 Code is that it doesn’t let you bring your own harness. When I tried it, it supported six harnesses, with Antigravity being one of them, which I personally think is useless.

For my Cursor subscription, using Orca means using the Agent CLI, which brings me back to the terminal issues I mentioned earlier. I can use it, but the experience isn’t great, especially when a thread gets long or I’m on my phone.

So while I prefer T3 Code’s UI, Orca gives me more flexibility with the agents and harnesses I want to run. That matters when I’m already paying for these subscriptions and want to use them.

T3 Code is also not a very stable product yet.

I’ve had it cut off my Cursor coding agents at various points after a lot of work was already done. That’s less of an issue now because I can restart them and pick up from where I left off, which wasn’t the case before.

Similarly, even with the claude integration, I get errors like “claude runtime stream failed” every other day. So yes, it’s not super stable yet. But picking up from where I left off is no longer a problem, and I’m willing to deal with those interruptions for now.

One other piece of feedback for T3 Code is about voice recording on mobile.

My phone is set to auto-lock after 30 seconds. If I’m rambling for longer than that while recording, my phone locks and I lose my transcription.

That has been my most annoying experience with T3 Code so far. I’d like it to keep the phone awake while I’m recording, or at least preserve the recording so I don’t lose what I just said when the phone locks itself.

T3 Code is going to be my daily driver for the next two or three weeks, and I’ll get back to you with more information.