Where do you talk to Amp most: terminal, editor, web, phone, Slack?
Ninety-nine percent of my usage is on the web, the rest through Slack because the context is already available there. I used to live in tmux and run everything in the terminal because tmux allows composing functionality like a diff viewer, editor, agents seamlessly with a keyboard focus. Orbs eliminated the need for multiplexing so I don’t spend time in the terminal anymore.
Favorite mode, and when do you switch?
I mostly use High mode as the default, Low mode when I know exactly what needs to be done and I just need a fast, intelligent typist, and Ultra for research problems like understanding bugs in distributed systems.
Do you still develop locally? If so, when?
Local development is now confined to cases that really need it, like sensitive production access.
What’s a workflow with Orbs that you absolutely love?
The coordinator/implementer pattern is great with orbs: for bigger tasks I have one Ultra or High orb that keeps the context, and delegates to sub-orbs for implementation. Yes, it sounds like a subagent, but orbs being isolated environments means that verification of individual stages is easier and you don’t risk cross-contamination of stages with throwaway scripts and other testing resources.
How do you review changes?
Code Review really should be Change Review because ultimately you are responsible for a running system, not code. Orbs are great for letting agents test changes end-to-end, construct ad-hoc interactive explainers, or capture telemetry traces before/after the change and review the diff of those.
Which skills, plugins, or MCPs do you use? Anything you built for yourself?
My Amp usage is pretty vanilla nowadays - most of the plugins and skills I consume or produce live in our repo because they are aimed squarely at the needs of our team. The Google Workspace MCP servers are genuinely useful because they make emails and documents much more accessible to the agent without needing to mess around with tricky API auth.
How do you use Puck?
I mainly use Puck through Slack, by tagging @amp on Slack.
What’s a prompt or phrase you keep reusing?
I like asking the agent “Would you deploy this?” or “Assess deployment impact” to tease out holes in the logic and potential oversights before deploying changes.
What do you still do by hand, and why?
Ultimately we are still responsible for what we ship, so I do test end-to-end by hand after a deployment.
Anything else? A surprising thing Amp did, a thing that didn’t work, a thing you wish it did.
For me verification is still the bottleneck, and the hard part is encoding quality or “taste” in a way that the agent can reason about.