Where do you talk to Amp most: terminal, editor, web, phone, Slack?
Web, phone is my second most used. I don’t use the TUI at all anymore, even over ssh… I prefer
using amp --no-tui and driving it from the web/phone.
Favorite mode, and when do you switch?
Medium, but right now high for GPT-6. I don’t switch except to use a new model.
Do you still develop locally? If so, when?
Yes. When I need to do GPU testing (GUI framework/wayland stuff). Even then I drive this from the web… I don’t really know that it’s running on my machine, I just need the GPU resource. Aside from this, I use my local runner to edit configuration files as needed.
What’s a workflow with Orbs that you absolutely love?
I have 2 laptops running at home with amp --no-tui with different GPUs. I have an orb working on
some wayland stuff and it needs to run GPU testing. It spawns a thread on a runner, runners do work
with whatever their resources are, and reports back to orb. Maybe they found we need to modify a
dependency, the orb spins up an orb for the dependency to make the change - dep messages orb back
that it’s done, orb updates, fans out to runners with specific resources, etc.
How do you review changes?
Video, screenshots, tables of data, “irrefutable proof”, vibe-check the diff size.
Which skills, plugins, or MCPs do you use? Anything you built for yourself?
- Skills: just one (outside what’s in the amp repo), a commit messages skill. Don’t really need it anymore.
- Plugins: I have a github webhook that gets events for anything on OSS repos of mine.
- MCP: none.
How do you use Puck?
I don’t use Puck. The few times I have was to do bulk archiving of threads, but now I am at steady state managing them so I don’t use Puck.
What’s a prompt or phrase you keep reusing?
(for whatever reason I don’t use the ship button)
What do you still do by hand, and why?
Prose, agent prose mostly sucks.
Anything else? A surprising thing Amp did, a thing that didn’t work, a thing you wish it did.
Discover other threads working concurrently, even when I haven’t asked it to. For example, I’ve had multiple threads spawned on a runner to do GPU testing - one of these threads discovered the GPU was in use, found the thread that had the GPU lock and then the threads began to coordinate shared usage.