@DiogoSnows There’s an equally expensive choke point: developers waiting on agents.
The useful interface isn’t another agent dashboard. It’s a clear handoff signal—working, blocked, or needs you.
Otherwise, managing the manager becomes another job.
@cursor_ai Agents as code is the right direction.
The next interface problem isn’t adding more agent controls. It’s reducing the attention tax: let the agent work, keep doing something else, and return only when the session actually needs a decision.
@wolfie_ Useful—but the /side command exposes the real problem.
Agent workflows are being stitched across panes, terminals, and tools.
The terminal should understand agent state and ask for attention—not make users build another layer just to watch it.
@lassejlv@mitchellh Impressive engineering. But rewriting Ghostty in another language still leaves the product question untouched:
What user work disappears?
If the answer is none, the terminal changed internally while the user's day stayed exactly the same.
@neogoose_btw Correctness matters. But even an honestly faster terminal can still solve the wrong problem.
If users still repeat logins, rebuild sessions, rerun commands, and babysit agents, speed only makes the waiting shorter.
It doesn't remove the work.
@fatwang2ai Three machines, different environments, multiple sessions—and the real problem isn't terminal speed. It's continuity.
The terminal should remember the hosts, sessions, credentials, and working context, so moving between machines doesn't mean rebuilding your day.
@dani_avila7 That switch says something important.
Ghostty proved speed matters. Rex shows speed was never the whole job.
Users don't spend their day admiring render latency. They manage sessions, panes, agents, and context. The terminal wins when those stop feeling like separate chores.
@Yuchenj_UW Desktop apps may be better for managing coding agents. That doesn't make the terminal the wrong interface.
It means the terminal shouldn't pretend to be an AI IDE.
Its job is shells, sessions, remote hosts, commands, and agent status—and doing that work exceptionally well.
@mitchellh Ghostty solved speed. What user problem did it solve after that?
You still manage hosts, repeat logins, rebuild sessions, rerun commands, and keep checking whether agents need you.
A faster terminal that leaves all the work intact is only a faster place to waste time.
@warpdotdev Human attention is the issue. But inviting agents into every workspace risks turning every workspace into another agent dashboard.
Our view: the terminal should not try to become Iron Man. It should work like JARVIS—quiet, prepared, and asking for attention only when needed.
Keep using what already works: OpenSSH, ~/.ssh/config, saved machines, local shells, tabs, and split panes.
XTTY. Stop working for your terminal. Build something that matters.
https://t.co/66Lldf4hKS
The terminal is not the hero. It should be quiet, prepared, and out of the way.
XTTY opens servers together, reaches production through jump hosts, fills recurring prompts from Keychain, restores duplicated remote tabs, broadcasts to selected panes, and shows agent status.
Stop judging terminals by what they demonstrate.
Judge them by what they eliminate.
Hosts. Logins. Sessions. Repeated commands. Agent checks.
The terminal is not the main character. Your work is.
https://t.co/MB5KW3Cey0
Do AI agents really need another terminal circus?
More dashboards. More panels. More commands. More attention stolen.
You need one thing: a clear status signal when the agent needs you. Then get back to the work.
I didn’t build xtty because terminals lacked features. I built it because the parts I use every day still felt separated—local shells, remote sessions, repeated operations, file transfer, and coding agents.
xtty brings them into one native macOS workflow.
https://t.co/EwJnLKWP3s
@missuo Good to see approval boundaries treated as a first-class feature. For agent-assisted server work, the hard part is showing exactly what will run, where it will run, and preserving context for review. How are you handling command context across multiple hosts?