If you’re new to @AmpCode I can’t recommend enough trying it as-is before layering in your own rules. It has extraordinarily good taste out of the box. We’re rolling out to more of our team and the majority of struggles I’ve paired w/ people through come down to oversteering.
brandRising
Amp on X
93 people say plainly that they love Amp and 23 credit Orbs with changing how they work; the complaints are a long tail rather than a shared grievance, and questions and wishes gather around Orbs.
First written 27 Oct · what changed · from 357 posts by 239 people
Tracking posts that say “@ampcode”
Track this for your marketPosts, last 24h
37
+517%vs. previous 24h
Posts, last 7 days
187
85 in the last 72h
Velocity
27/day
7-day average
Momentum
+39%
Last 24h vs. a normal day this week
Posts per day, last 30 days
as of Nov 3, 10:00 UTCWhat changed
as of 27 Oct · checked once a monthFirst written 27 Oct. When what people say changes, it shows here.
Tone· 193 posts from 143 people· written 27 Oct
How do people feel about Amp?
Most of the talk is plain enthusiasm — 93 people say they love using it — and Orbs are what they credit with changing how they work (23 people). Criticism is a small tail: 8 people came away disappointed.
Praise is specific rather than generic. People point to the agent's judgement out of the box ("extraordinarily good taste out of the box"), to before/after screenshots and inline windows it opens so work can be checked inside the chat, to it noting CI checks and PR review bots, and to the mobile app and webhooks. One person describes talking to it through a project they had been putting off.
Orbs carry the strongest claims: a self-hosted setup on Firecracker microVMs where deleting an orb takes the files, checkout and processes with it; end-to-end testing on Linux, Windows and Mac with videos sent back; a local runner and Orb threads passing files between them.
The 8 disappointed call it "overly engineered", find the UI complex and the spread of models at odds with its "focus on the task" promise, and note that cloud-only work stops when the connection does. 19 are undecided, mostly asking what to start with.
- Loves it126
- Orbs won them over26
- Undecided20
- Ahead of the pack12
- Disappointed9
See 5 of 193 postsHide the posts
ok, and @AmpCode self hosted orbs, on my @omarchy mac pro w/ 512gb ram (using firecracker microvms) seems to have solved so many problems still early, but working well i can smash the 'ship' button and call it a day. when orbs get deleted - all files, the checkout, the running processes, everything - is gone no more filling up disk, no more memory leaks, no more orphaned processes, no more branches and worktrees that don't get cleaned up all orbs are running on my tailnet too, which is handy oh, and they all are configured with a cliproxyapi instance load balancing 3x claude subscriptions. i'm just bounded by my ability to ideate and design
This is why orbs are powerful: @AmpCode could test this feature end to end on Linux, Windows, and Mac, and was sending me regular videos of how things work. My work was thinking at high level through all the "what if..." scenarios and guiding the agent toward a solution.
As a Dart Flutter developer, I need to test the debug build. Only @AmpCode can do that as of now; other harnesses struggle to connect with the debug build and execute taps and clicks.
So far, honestly it feels overly engineered and over complex. And I'm still struggling to see why it's better than my agents setup with subagents and an orchestrator running them
5 of the 193 posts behind this answer, one per person. The other 188, and every new one as it’s posted, come with tracking. Get early access
Pain points· 82 posts from 60 people· written 27 Oct
What do people complain about with Amp?
Bugs and rough edges are the largest group, named by 19 of 60 people, ahead of cost (11) and outages (7). No theme here accounts for half the people; the gripes are many and scattered.
The bug reports are small and varied: collapsed threads re-opening on their own, a thread not picking the Claude Code model when spinning up an Orb, voice taking a long time to connect on iOS, markdown diagrams not rendering, a balance showing $0.88 in one place and $0 in another, and bug reports that "seem to go into a black hole".
Cost complaints centre on usage being hard to predict — credits consumed faster than doing the same work locally, a monthly figure quoted against API credits, and a self-hoster saying the difference against a local subscription is "insane".
Outages are reported as whole lost days: cloud agents unreachable, Orbs broken for half a day. One person's line is the trade-off in miniature: shipping fast is not welcome at the expense of reliability.
5 people cannot tell whether a thread draws on their subscription or API tokens; the answer they pass around is that picking Claude in the dial bills API prices. 6 hit subscription and rate limits, and 5 find Orbs hard to fit to large data or to orchestrate.
- Bugs and rough edges22
- Costs too much11
- Outages and downtime11
- Confusing model dial and billing9
- Orbs hard to use7
- Subscription limits6
See 5 of 82 postsHide the posts
@AmpCode minor annoyance started some time today I believe. Collapsed threads re-open without me triggering it. Makes the sidebar unwieldy and annoying to navigate. 🙏
I have been playing with @AmpCode a few days now connected with my Claude subscription and I noticed that credits are consumed much faster than if I was doing the same thing locally. I guess going through the Claude Code SDK that Amp is doing doesn't get the same discounts...
These @AmpCode downtimes started to get annoying. Shipping features fast is not welcomed, if it is at the expense of reliability.
picking claude in the dial means you are paying API prices and using up your Amp usage, not your sub.
I suspect people who find orbs to be extremely powerful/effective might work on a small number of repos and possibly on specific kinds of products where orbs make sense. I've found trying to orchestrate orbs like I used to orchestrate w/ solo to be very difficult.
5 of the 82 posts behind this answer, one per person. The other 77, and every new one as it’s posted, come with tracking. Get early access
Alternatives· 26 posts from 25 people· written 27 Oct
What are the alternatives to Amp?
The field is wide and even: T3 Code and Pi are level, each named by 5 and 5 people, with OpenCode, Claude Code and Cursor level behind them at 3. The comparison is usually about where work runs, not which agent is smarter.
People describe running things side by side rather than picking a winner: one splits projects 50/50 between Amp and T3 Code, using Claude models through T3 and OpenAI and Grok subscriptions through Amp; another gives a bot Claude Code, Codex and Amp and hands work to whichever fits. T3 Code comes up as the cloud-centric alternative that keeps CLI, desktop, web and mobile features in step; OpenCode and Claude Code as things used in the same flow.
Cursor is weighed over money: one developer says the inability to use an OpenAI or Claude subscription there is the blocker, against using those subscriptions inside Amp.
Orbs are the usual answer to Conductor's remote workspaces and to the question of what would make someone switch from omp — with one person asking why they should switch when they can already use their own subscription in Conductor. Devin, Droid and Codex come up once each; one person says they go back to Droid when they lack an option.
- T3 Code5
- Pi5
- OpenCode3
- Claude Code3
- Cursor3
- OMP2
See 5 of 26 postsHide the posts
Já teď jedu tak 50/50 @AmpCode a na starším compu jsem nainstaloval serverové Ubuntu a tam mám projekty které jedu vzdáleně přes @t3dotcodes U Amp mě štve jediná věc, a to že přes předplatné tam jde jen OpenAI a Grok, a přes T3 jedu zase hlavně Claude modely. Přes API cokoliv.
Micky and I came to the same revelation on the lord's day. Must be devine 😂 I gave my Grok Bot three coding agents today: @claudeai Code, @ChatGPT Codex and @AmpCode. Why: my @bot is great at planning and checking work, but heavy coding ate my weekly usage. Now it hands the
not being able to use our @OpenAI or @claudeai subscriptions is @cursor_ai is the biggest blocker for me like no way am i going to spend $4,000 usd on API credits through cursor when I can use my subscription in @AmpCode for $300/mo pls @poteto work your magic
Why are orbs better then conductor remote workspaces? I can use my own subscription in conductor. Tell me why I should switch?
That said, probably @ampcode or @t3dotcodes because they are cloud-centric at their core so the native cli/gui/webui/mobile all have nearly the same features. Whereas Codex/Claude/etc have disjoint featuresets depending on your entrypoint.
5 of the 26 posts behind this answer, one per person. The other 21, and every new one as it’s posted, come with tracking. Get early access
Questions· 51 posts from 43 people· written 27 Oct
What do people ask about Amp?
Self-hosting Orbs is the largest question, asked by 11 people, just ahead of 10 asking what Amp actually does that other agents don't. 8 want to know what Orbs are and whether everything runs in the cloud.
The self-hosting questions follow one person's detailed post about running self-hosted Orbs on a Mac with Firecracker microVMs: how hard setup was, whether they can run on their own VPS, whether they use their own image or the Amp-provided one, and how Orb startup compares with Amp-hosted Orbs. Others ask when to use Orbs rather than self-hosted runners; the answer given is runners for host access, Orbs for isolation.
The "what makes it different" questions are blunt: what is it shipping the rest aren't, why is it better than a simple loop over a model, how does it compare with Cursor's and Claude's cloud agents. A few ask for a learning resource or a tutorial.
7 ask where a feature lives — the Claude Code option turned out to be in the new-thread dial, not the build dial. 4 ask where the Oracle and the model dial went after they disappeared, and 3 ask whether a Claude subscription can be used or usage has to be paid.
- Can I self-host Orbs?13
- What makes Amp different?10
- What exactly are Orbs?8
- Where do I find this feature?7
- Where did the Oracle go?5
- Can I use my Claude subscription?3
See 5 of 51 postsHide the posts
Nice! How easy/hard was setup for self-hosted orbs?
Can you run them on your own vps?
appreciate it! I've seen Orbs everywhere, are they different from general cloud coding agents? Or maybe the UX is much better? GOing to give it a try
"Not as AGI pilled" is doing a lot of work here. What is AmpCode actually shipping that the rest aren't — or is the gap just velocity and a team that dogfoods its own product?
I checked on amp code app and web.. in the build dial section, I tired to search for Claude Code and couldn’t see it.
5 of the 51 posts behind this answer, one per person. The other 46, and every new one as it’s posted, come with tracking. Get early access
Wishes· 27 posts from 24 people· written 27 Oct
What do people want from Amp?
The wishes are scattered and small: an @ mention picker for bringing agents, tickets and threads into a prompt (2 people, the largest theme, though they are 5 posts from few voices), and multiple subscriptions used side by side (4). None of the wants is shared by many people.
The @ mention discussion is concentrated: one person wants a picker that can pull in Linear tickets, MCP connections, other threads and other people, and points at Grokbot's version; another misses typing file paths from the app; a third argues people still thinking in filenames are not letting go enough.
Multiple subscriptions is asked plainly: can several OpenAI and Claude accounts be used, with one person spelling out the wish for an Oracle model on a ChatGPT subscription and the main agent on a Claude one.
Two people want the Oracle back — "cross-model code review was my main reason for using Amp Code" — and two more want to specify which model subagents use, which one answer says is now a plugin rather than a built-in dial. Self-hosters ask to pass Orb size through to a single provider rather than one provider per size.
Single asks: Cursor-style Projects for managing a stream of work, frictionless voice input, an Android app, portals for self-hosted runners, splitting a thread into parallel work, a native ARM64 build image, and the Orb size picker back.
- @ mention for agents and context5
- Multiple subscriptions at once4
- Paid Oracle/Librarian tiers2
- Bring back the Oracle2
- Specify the subagent model2
- Pass orb size to providers2
See 5 of 27 postsHide the posts
With how much coordination between agents happens these days, I'm kinda itching for an @ mention tool for this @AmpCode what do you think @sqs ?
I keep wanting to do that. Best case is making the threads by task topic or runner. Seems to go ok, but feels like it's mostly out of luck. I also miss being able to at-file paths from the apps.
+1 to this. Would be cool to have multiple subs cover different modes or agents! E.g. Astra Oracle via ChatGPT sub with Claude sub for main agent.
Dear @AmpCode, please bring back the Oracle! 🙏 Cross-model code review was my main reason for using Amp Code. E.g. Claude as agent (implementer) and chatGPT as Oracle (reviewer) and/or vice versa.
At the very least give us the option to specify a subagent model.
5 of the 27 posts behind this answer, one per person. The other 22, and every new one as it’s posted, come with tracking. Get early access
In their words
How do people talk about Amp?
The phrases people use most, counted once per person.
- 1“claude sub”5
- 2“claude subscription”4
- 3“cloud agents”4
- 4“self-hosted orbs”3
- 5“local runner”3
- 6“hosted orbs”3
- 7“model dial”2
- 8“low cache”2
- 9“reasoning effort”2
- 10“remote agents”2
- 11“code review”2
- 12“mac orbs”2
Posts are shown the way X shows them. One of them yours? Here’s why it’s shown, and how to take it off.
Also in AI coding agents and editors
See all 12 in AI coding agents and editors →- Cursor+41%People still credit Cursor with real shipped work and recommend it by name, but the argument has moved to what the bundled X plan gives them and what happens when something breaks.1,392 posts / 7d
- Claude Code+81%The biggest single theme is finished work: people describe apps, automations and fixes Claude Code completed, often naming the setup that made it work.915 posts / 7d
- OpenAI Codex+36%The biggest share of the talk is people showing what Codex builds for them.873 posts / 7d
- Cognition (Devin)+8%Cognition's talk centres on a lead-plus-sidekick setup people read as cheaper per task at the same score, and on firsthand users sold on cloud agents — set against crashes, lost sessions, price and limits, with Claude Code the comparison people keep making.557 posts / 7d
Get every post about Amp, filtered and labelled
This page is the summary. Tracking gives you the stream behind it for any brand, keyword or @mention: each new pain point, request and question as it’s posted, with the spam and promotions already taken out.