# Amp (Amp)

Feedback Bench, coding agents, built 2026-10-19, window 2026-08-31 to 2026-10-18. Web page: https://feedbackbench.com/coding-agents/#/agent/amp

| Measure | Value |
|---|---|
| Rank | 5 of 17 (rank range 5–6) |
| Feedback Score | 56.0 (95% interval 54.2–57.8) |
| Popularity | 0.482 (share of voice 1.02%) |
| Customer love | 0.578 (95% interval 0.563–0.593) |
| Place | Popularity 11th of 17 × Customer love 3rd of 17 |
| Top quadrant | no |
| Users | 1398 |
| Posts counted | 5118 |
| Posts that judge the agent | 1831 |
| Criteria ranking high / low | 11 / 1 of 63 |

## The brief

Written by large language models from this page's numbers and the posts it quotes, then checked: every number in the text is in the build. Interpretation, not measurement.

Amp asks users to learn a new way of working, and those who do mostly like the result. Users run coding sessions on cloud machines, steer them from a phone, and praise the work that comes back. The interface makes up 34% of what its users rate, 3.0 times the rate for other agents. The same surface also pulls it down. Users find the display and settings hard to follow, and they raise this 2.4 times as often as users of other agents. Amp ranks 5th of 17 overall.

### Why people choose it

- **Cloud sessions you can steer from a phone** ([Interface](https://feedbackbench.com/coding-agents/criteria/interface.md)). 3.0×: 34% of what users rate, 3.0× the rate for other agents. Ranks 2nd of 14 on “Interface”. Posts describe starting and steering work from an iPhone or iPad while the code runs on a dedicated cloud machine for each thread. Some users say they have moved most of their development off their laptop. Others say that keeping the app they are building running in the cloud is a draw.
  Evidence:
  - 2026-10-17, @AmpCode (X): “@d4m1n my guess is that i already do around 20% of my product building from my iphone, @ampcode is the killer iphone duo app, like evernote was for the first ipad” [source](https://twitter.com/14405744/status/2111430017448058963)
  - 2026-10-14, @AmpCode (X): “first orb experience on @ampcode. so interesting how it's like the "under the hood" version of grok bot/muse, with a dedicated vm per thread. having the app i'm developing running there vs on my machine is compelling.” [source](https://twitter.com/10079052/status/2110455715022376970)
- **Threads and subagents that hand work along** ([Quality of work](https://feedbackbench.com/coding-agents/criteria/work.md)). 2nd: of 14 on “Quality of work”. 26% of what users rate. Users describe agents that open new threads with the right context, message the original thread, watch CI and move pull requests toward ready. Teams like multiplayer threads, where several people work with one agent, and say these cut down on Slack posts.
  Evidence:
  - 2026-10-15, @AmpCode (X): “@weswinder i love @ampcode because claude can just make a new thread for me and give it all the context and the new thread can, if needed, read the latest messages (not the full verbose history with tools but messages) and even message the og thread, it’s so good” [source](https://twitter.com/2088622190925135872/status/2110878423522623751)
  - 2026-10-18, @AmpCode (X): “@marccampbell the intergration @ampcode added recently is amazing. it watches for ci and reviews and automatically takes actions to get move the pr to a ready state.” [source](https://twitter.com/5444392/status/2111641515269501385)
- **Your Claude and OpenAI subscriptions work inside it** ([Pricing and limits](https://feedbackbench.com/coding-agents/criteria/paying.md)). 4th: of 13 on “Pricing and limits”. 24% of what users rate. Users welcome being able to bring their existing Claude and OpenAI subscriptions into Amp. Some say they moved most of their work there because of it. Others say their usage lasted longer than they expected.
  Evidence:
  - 2026-10-10, @AmpCode (X): “for all the people asking if you can use your claude plan in @ampcode - now you can! i had early access to this and it’s allowed me to move basically all of my work to amp and orbs with my claude and openai subs. [link]” [source](https://twitter.com/9111552/status/2108814443467645235)
  - 2026-10-17, @AmpCode (X): “i must have done something stupid the first time i tried orbs a month ago. my orbs usage disappeared in 2 days. i've been using it for 2 days now and still have 98% left. loving @ampcode” [source](https://twitter.com/568738762/status/2111518875489861993)

### Why people leave

- **Display and settings are hard to follow** ([Clear display and settings](https://feedbackbench.com/coding-agents/criteria/ui.display_settings.md)). 2.4×: 15% of what users rate, 2.4× the rate for other agents. Ranks 8th of 12 on “Clear display and settings”. Users call the interface complex and opinionated. They say settings sit in unexpected places, such as model choice on a dial. Posts describe collapsed items that keep reopening, especially on iPhone, and pending messages that are easy to miss.
  Evidence:
  - 2026-10-13, @AmpCode (X): “@sqs @ampcode the ui feels complex, maybe i just need to adjust to it. i've been using [link] for months. also have used claude code and cursor off and on for like 2ish years since the early days” [source](https://twitter.com/1232338194/status/2109857021843697746)
  - 2026-10-10, @AmpCode (X): “@sqs @ampcode this is still happening. collapsed items reopening all the time. seems worse on mobile (iphone) can you take a look” [source](https://twitter.com/33135576/status/2108989121373089911)

### What would move its score

Rule: For each area where an agent's customer love is below the category norm: the Feedback Score and rank it would have if that one area were at the norm (0.5) and everything else stayed the same, computed with the same formula as the ranking. An area's pull follows its weight, so an area many users of the category rate moves the score more than one they rarely mention.

Moving any one area to the category norm would not change the Feedback Score, because every area is already at or above it. The weak spots users name, display and settings and complaints about errors and crashes, sit inside areas whose overall ratings already meet the norm.

### Six weeks

Weekly users rose from 270 to 398 over the window, while users across all coding agents fell from 32,348 to 26,760. Share of voice held near 1%. Rank moved between 5th and 7th and finished at 5th.

## Top requests

What users ask to add or change, most asked first. 468 user-weeks ask for something. Requests do not change the Feedback Score. Rule: A separate pass by a large language model reads every counted post and extracts what the user asks the agent or its vendor to add or change, with the criteria it maps to and a short normalised wording; it does not touch the labels or the Feedback Score. A large language model groups the wordings within each criterion (the first criterion the request maps to) into themes, and a second pass merges themes that ask for the same change under different criteria; a merged theme sits under the criterion most of its requests map to, and code counts it once. A theme counts distinct user-weeks that ask for it, per agent; across agents, one user-week per agent. Themes asked in fewer than 2 user-weeks, and requests that share no theme, are not shown. Examples: up to 3 posts per theme from different users, without slurs: whole posts of 60 to 450 characters with at least 8 words beyond handles and links, posts of at most 280 characters first, then the newest.

| Rank | Request | Criterion | User-weeks | Posts |
|---|---|---|---|---|
| 1 | Claude subscription usable in third-party harnesses | [Using an existing subscription across tools](https://feedbackbench.com/coding-agents/criteria/billing.subscription_portability.md) | 35 | 38 |
| 2 | Subscriptions usable in third-party harnesses | [Using an existing subscription across tools](https://feedbackbench.com/coding-agents/criteria/billing.subscription_portability.md) | 17 | 19 |
| 3 | Multiple subscriptions active at once | [Using an existing subscription across tools](https://feedbackbench.com/coding-agents/criteria/billing.subscription_portability.md) | 8 | 10 |
| 4 | Run cloud agents on own infrastructure | [Cloud and remote sandbox execution](https://feedbackbench.com/coding-agents/criteria/surfaces.cloud_sessions.md) | 8 | 8 |
| 5 | macOS cloud environments for agents | [Cloud and remote sandbox execution](https://feedbackbench.com/coding-agents/criteria/surfaces.cloud_sessions.md) | 7 | 8 |
| 6 | Group sidebar sessions by project | [How the interface shows work, and what the user can configure](https://feedbackbench.com/coding-agents/criteria/ui.display_settings.md) | 6 | 7 |
| 7 | Customizable model picker | [How the interface shows work, and what the user can configure](https://feedbackbench.com/coding-agents/criteria/ui.display_settings.md) | 5 | 6 |
| 8 | Custom OpenAI-compatible provider endpoints | [Connecting own API keys, local models and custom endpoints](https://feedbackbench.com/coding-agents/criteria/setup.provider_byok_local.md) | 5 | 5 |
| 9 | GPT-6 Sol and Luna model support | [Which models are offered on a plan and when](https://feedbackbench.com/coding-agents/criteria/models.catalog_access.md) | 5 | 5 |
| 10 | Linux desktop app and Linux support | [Install, launch and sign-in](https://feedbackbench.com/coding-agents/criteria/setup.install_signin.md) | 5 | 5 |
| 11 | Multi-account switching | [Install, launch and sign-in](https://feedbackbench.com/coding-agents/criteria/setup.install_signin.md) | 5 | 5 |
| 12 | Opus 5.5 model support | [Which models are offered on a plan and when](https://feedbackbench.com/coding-agents/criteria/models.catalog_access.md) | 5 | 5 |

### 1. Claude subscription usable in third-party harnesses

- Amp, 2026-10-10, @AmpCode (X): “@adamwathan @ampcode the barrier is always the same. can i bring claude sub or not.” [source](https://twitter.com/2840622061/status/2108734680162087252)
- Amp, 2026-10-04, @AmpCode (X): “@nikstep @ampcode i came away super impressed. got set up in like 5 minutes. but tbh if i can’t use subsidized anthropic models it will have to wait. it’s really unfortunate. [link]” [source](https://twitter.com/2544193734/status/2106810124375392513)
- Amp, 2026-09-30, @AmpCode (X): “would be great if we could use @ampcode orbs with our @claudeai subscription” [source](https://twitter.com/1743576399191883776/status/2105251675208618446)

### 2. Subscriptions usable in third-party harnesses

- Amp, 2026-10-15, @AmpCode (X): “@ampcode maybe interested in supporting the usecase "bringing in my copilot subscription"? employer pays the sub and has a whole bunch of models.” [source](https://twitter.com/2567249282/status/2110620846796554245)
- Amp, 2026-10-13, @AmpCode (X): “@ampcode your amp oauth page doesn't have "sign in with codex" option. so if we signed up to your service with it and have access to the web app. but then try to connect the cli to it, it doesn't work.” [source](https://twitter.com/15952889/status/2109866051681112419)
- Amp, 2026-10-13, @AmpCode (X): “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” [source](https://twitter.com/1788108425400123392/status/2109845059835412587)

### 3. Multiple subscriptions active at once

- Amp, 2026-10-14, @AmpCode (X): “@jeremylangevin9 @ampcode @thorstenball @sqs +1, i wouldn't mind this as well. or maybe subscription selector within the dial itself” [source](https://twitter.com/2000470921329709056/status/2110301031821099435)
- Amp, 2026-10-12, @AmpCode (X): “@sqs @ampcode really appreciate your active support here on x 1. understood, make sense. and there is also no "workaround" like a "community plugin system" ? 😂 2. would be amazing to use dial with the subs” [source](https://twitter.com/3352891385/status/2109749435269869609)
- Amp, 2026-10-12, @AmpCode (X): “i love that @ampcode now supports claude! one request, @thorstenball and @sqs is there any chance that project-based connections is a thing coming soon? (work vs play ai subs based on project settings).” [source](https://twitter.com/1405835361021042691/status/2109653765561504200)

### 4. Run cloud agents on own infrastructure

- Amp, 2026-10-14, @AmpCode (X): “@ampcode finally gave orbs a proper try and this really is fantastic. is there any way to run the orb on a machine i control? (i have a quite powerful nas that spends most of its life unused)” [source](https://twitter.com/1007894812744282112/status/2110251032471343296)
- Amp, 2026-10-13, @AmpCode (X): “@karlclement @flodev @sqs @tobimori @ampcode @sagtanih would this be something we can jump on as well? would love to beta test byo orbs.” [source](https://twitter.com/1907125052736184321/status/2109884369846370439)
- Amp, 2026-10-13, @AmpCode (X): “@ampcode portals for self hosted runners wen 😂 desktop mode works but alas i miss my cloud features.” [source](https://twitter.com/22610703/status/2109814075274522804)

### 5. macOS cloud environments for agents

- Amp, 2026-10-12, @AmpCode (X): “@taylorotwell @ampcode until morbs are available everywhere, ios development definitely needs some custom runners.” [source](https://twitter.com/20612511/status/2109658491405373693)
- Amp, 2026-09-30, @AmpCode (X): “dear @ampcode , any plans to release mac orbs? that paired with a could ios device would make ios dev so much easier.” [source](https://twitter.com/881577286927036416/status/2105226945461932208)
- Amp, 2026-09-25, @AmpCode (X): “@sqs @sixhobbits @ampcode @sqs they’re good. i really liked devin macos orb. amp macos orb wen?” [source](https://twitter.com/1553373273639256065/status/2103437963472539744)

### 6. Group sidebar sessions by project

- Amp, 2026-10-18, @AmpCode (X): “@sqs @crasperjoome @ampcode having a better sidebar would help a lot!” [source](https://twitter.com/474153793/status/2111888285324693551)
- Amp, 2026-09-07, @AmpCode (X): “@sqs @josevalerio @ampcode add an agent-controlled sidebar mode to let a subagent present the best "current" threads tree/list/etc” [source](https://twitter.com/132882990/status/2096911111131894250)
- Amp, 2026-09-06, @AmpCode (X): “i beg you @ampcode i don’t want to make projects, or tag shit manually give me the clusters of work again [link]” [source](https://twitter.com/1420340648117510149/status/2096739242193887449)

### 7. Customizable model picker

- Amp, 2026-09-26, @AmpCode (X): “@ampcode it would be nice if model selection was ordered by used most often. i often find myself switching between a couple of models but having to scroll through all of the options. this in all uis” [source](https://twitter.com/188892343/status/2103868071928897983)
- Amp, 2026-09-23, @AmpCode (X): “@jkudish @ampcode agree! could be an option in "build your own dial" to just add a couple more. sometimes i want "high but with fable" vs my regular high with sol.” [source](https://twitter.com/1567548474794835969/status/2102856772021305745)
- Amp, 2026-09-23, @AmpCode (X): “@skastr052 @ampcode maybe with build your own dial, you could choose how many spots you have up to a reasonable maximum?” [source](https://twitter.com/33135576/status/2102841110309797938)

### 8. Custom OpenAI-compatible provider endpoints

- Amp, 2026-09-14, @AmpCode (X): “@benvargas @ampcode @sqs same! have been waiting a very, very long time to connect my cliproxyapi… [link]” [source](https://twitter.com/1455730994976227331/status/2099510909400195253)
- Amp, 2026-09-07, @AmpCode (X): “when will the megawatt subcription account be enabled for custom base url byok?, i’ve been waiting for it so long @sqs @ampcode 🥲” [source](https://twitter.com/1800425844491620352/status/2096782766776221869)
- Amp, 2026-09-01, @AmpCode (X): “@sqs @socksmyrocks @ampcode would it add support for any baseurl or only whitelisted?” [source](https://twitter.com/817707228912357377/status/2094878138320695803)

### 9. GPT-6 Sol and Luna model support

- Amp, 2026-09-28, @AmpCode (X): “why still no gpt6-sol in the "low" and "medium" dial modes? @ampcode [link]” [source](https://twitter.com/2062509956574650368/status/2104601900461215997)
- Amp, 2026-09-22, @AmpCode (X): “i'm seeing visions of 6 sol agent + opus 5.5 oracle and i need it yesterday in @ampcode” [source](https://twitter.com/1281249457703538688/status/2102474111570366753)

### 10. Linux desktop app and Linux support

- Amp, 2026-10-17, @AmpCode (X): “@rockorager @ampcode we need this native omarchy amp! match made in heaven” [source](https://twitter.com/1693542458997563392/status/2111510416484393321)
- Amp, 2026-10-16, @AmpCode (X): “@rockorager @ampcode where can i get the native version that runs on linux? the macos app has been pretty good, but i miss being on linux!” [source](https://twitter.com/129716485/status/2111201625788301591)
- Amp, 2026-09-30, @AmpCode (X): “hey @ampcode, any chance ya boy can get a linux app? i got used to having the app on my macbook but my home machines are ubuntu.” [source](https://twitter.com/1068569182986334208/status/2105381827800162769)

### 11. Multi-account switching

- Amp, 2026-10-15, @AmpCode (X): “if @ampcode supported multiple accounts for each sub-provider, i would ditch it for everything else. it's such a great product, and orbs are very transparently implemented - i never really have to care about them. but currently having to login/logout is a pain. t3code nails it” [source](https://twitter.com/165356795/status/2110628837541474610)
- Amp, 2026-10-10, @AmpCode (X): “@sqs @ampcode can we use multiple openai and claude accounts?” [source](https://twitter.com/1004220018/status/2108792049789272138)
- Amp, 2026-09-28, @AmpCode (X): “@sqs @homborg @ampcode i've tried a several times to sign into both of the accounts in this screenshots, but i only see one of them in amp. i'm able to use both accounts with the codex cli. i switch between them when i run out of usage. [link]” [source](https://twitter.com/36411940/status/2104539048916156631)

### 12. Opus 5.5 model support

- Amp, 2026-10-10, @AmpCode (X): “@benvargas @____hnsn @ampcode yeah that's my medium mode right now and add opus 5.5 medium for subagents :)” [source](https://twitter.com/1705384263867379712/status/2108794699133903047)
- Amp, 2026-09-23, @AmpCode (X): “we need a strong response with astra 6.1. still playing with opus 5.5. hammering all night didn't move any usage much. not super impressed but it's moving much smarter than gpt 6 sol. biggest pain point is not driving via @ampcode for opus 5.5. [link]” [source](https://twitter.com/51510563/status/2102771724643168704)
- Amp, 2026-09-22, @AmpCode (X): “@ampcode it's been sometime - looking forward to your post about opus 5.5 and update to the default dial :p” [source](https://twitter.com/85549810/status/2102480563596591364)

## Ranks highest on

Support, refunds and issue handling (1st of 8), Can do the user's kind of task (5th of 13), Subagents, parallel agents and orchestrators (1st of 10)

## Ranks lowest on

How the interface shows work, and what the user can configure (8th of 12)

Position on a criterion: its place among the agents with at least 30 rated user-weeks there, ordered by customer love. It ranks high when both its position and its 95% interval put it in the top half, low when both put it in the bottom half; neither otherwise, or when fewer than 4 agents qualify.

## All 63 criteria

Criterion love: 0.5 is the category norm. n: rated user-weeks.

### Paying and limits: 4th of 13 (customer love 0.587, n 222)

Users talk about using their Claude and OpenAI subscriptions in Amp. This is 10.6% of what they rate, against 1.1% for other agents, and views on it are split. Complaints also center on usage quota running out, which draws 30 complaints.

| Criterion | Reading | Customer love | 95% interval | n | Praise | Complaint |
|---|---|---|---|---|---|---|
| [Using an existing subscription across tools](https://feedbackbench.com/coding-agents/criteria/billing.subscription_portability.md) | 2nd of 8 | 0.528 | 0.501–0.559 | 97 | 47 | 50 |
| [How much use a plan's price buys](https://feedbackbench.com/coding-agents/criteria/limits.plan_value.md) | 6th of 12 | 0.527 | 0.502–0.549 | 66 | 35 | 31 |
| [Single prompt, model or effort level consumes disproportionate quota](https://feedbackbench.com/coding-agents/criteria/limits.burn_rate.md) | 6th of 12 | 0.503 | 0.482–0.527 | 37 | 7 | 30 |
| [Usage meter visibility and accuracy](https://feedbackbench.com/coding-agents/criteria/limits.usage_meter.md) | Too few posts | 0.525 | 0.490–0.564 | 24 | 5 | 19 |
| [Free tier and free model availability and limits](https://feedbackbench.com/coding-agents/criteria/billing.free_tier.md) | Too few posts | 0.508 | 0.498–0.519 | 12 | 9 | 3 |
| [Short rolling usage window blocks or interrupts work](https://feedbackbench.com/coding-agents/criteria/limits.window_interrupts_work.md) | Too few posts | 0.496 | 0.488–0.506 | 8 | 1 | 7 |
| [Price, allowance or plan terms changed](https://feedbackbench.com/coding-agents/criteria/limits.allowance_change.md) | Too few posts | 0.512 | 0.490–0.550 | 8 | 1 | 7 |
| [Quota reset timing and bonus or banked resets](https://feedbackbench.com/coding-agents/criteria/limits.reset_schedule.md) | Too few posts | 0.508 | 0.495–0.525 | 7 | 3 | 4 |
| [Pay-as-you-go overage, fallback billing and spend caps](https://feedbackbench.com/coding-agents/criteria/billing.overage_charges.md) | Too few posts | 0.497 | 0.494–0.500 | 4 | 0 | 4 |
| [Pricing and plan terms stated clearly and consistently](https://feedbackbench.com/coding-agents/criteria/billing.pricing_clarity.md) | Too few posts | 0.498 | 0.494–0.500 | 3 | 0 | 3 |
| [Prompt cache hits, misses and invalidation](https://feedbackbench.com/coding-agents/criteria/limits.prompt_cache.md) | Too few posts | 0.498 | 0.494–0.500 | 2 | 0 | 2 |

Most recent posts:

- Praise, 2026-10-17, @AmpCode (X): “i must have done something stupid the first time i tried orbs a month ago. my orbs usage disappeared in 2 days. i've been using it for 2 days now and still have 98% left. loving @ampcode” [source](https://twitter.com/568738762/status/2111518875489861993)
- Praise, 2026-10-15, @AmpCode (X): “when subscribing to @ampcode's $20 plan, i was afraid usage bar anxiety wouldn't let me do my orbin' calmly. three days later, the feeling is i don't do enough orbin'. and i'm not cringing anymore at calling it orbin'. it's just what it is.” [source](https://twitter.com/1973351797482786816/status/2110878563763413259)
- Praise, 2026-10-13, @AmpCode (X): “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” [source](https://twitter.com/1788108425400123392/status/2109845059835412587)
- Complaint, 2026-10-14, @AmpCode (X): “amp gives me a new coding experience, but why does it still incur costs when i use my own subscription and runner? @ampcode [link]” [source](https://twitter.com/2293726776/status/2110193950162383317)
- Complaint, 2026-10-10, @AmpCode (X): “@ampcode @sqs ampcode says i’ve hit my chatgpt limit, but i still have 54% left for the week. i’ve logged out and back in twice [link]” [source](https://twitter.com/757968450291544065/status/2108774522140897470)

### Setting up and connecting: 3rd of 12 (customer love 0.525, n 170)

Users talk about setup more than users of other agents do. Install and sign-in ranks 1st of 8 but still draws more complaints than praise. Views on MCP plugins and on using your own keys or local models are split about evenly.

| Criterion | Reading | Customer love | 95% interval | n | Praise | Complaint |
|---|---|---|---|---|---|---|
| [MCP servers, plugins, skills and hooks](https://feedbackbench.com/coding-agents/criteria/setup.extensions_mcp.md) | 3rd of 9 | 0.504 | 0.485–0.523 | 49 | 25 | 24 |
| [Connecting own API keys, local models and custom endpoints](https://feedbackbench.com/coding-agents/criteria/setup.provider_byok_local.md) | 5th of 9 | 0.507 | 0.487–0.526 | 48 | 27 | 21 |
| [Install, launch and sign-in](https://feedbackbench.com/coding-agents/criteria/setup.install_signin.md) | 1st of 8 | 0.540 | 0.507–0.573 | 44 | 15 | 29 |
| [Onboarding, discoverability and documentation](https://feedbackbench.com/coding-agents/criteria/setup.onboarding_docs.md) | 2nd of 7 | 0.521 | 0.497–0.547 | 33 | 11 | 22 |
| [IDE and editor integration](https://feedbackbench.com/coding-agents/criteria/setup.ide_integration.md) | Too few posts | 0.500 | 0.490–0.509 | 10 | 5 | 5 |

Most recent posts:

- Praise, 2026-10-13, @AmpCode (X): “puck don't f* around so simple to set up custom endpoints on @ampcode ps. that "secure mask dialog" ui component 🤌🏻👏 [link]” [source](https://twitter.com/3159431/status/2109806983113007120)
- Praise, 2026-10-13, @AmpCode (X): “@maxmax @chatgpt @ampcode watching a dot just plug into amp and start working feels like magic that actually works. the simplicity is wild, no setup, no friction, just an always-on teammate impressed by how smooth it is.” [source](https://twitter.com/1498556764391755776/status/2109811328558969222)
- Praise, 2026-10-13, @AmpCode (X): “having [link] added to codex (or any other mcp client) allows passing work to @ampcode to anywhere - super useful! the feature is still marked experimental; some issues are possible. just tried it with codex - worked like a charm!” [source](https://twitter.com/103563938/status/2109942267100451174)
- Complaint, 2026-10-18, @AmpCode (X): “@ampcode sign-in on the macos app isn't working... after completing authentication in the browser, the redirect back to the app fails. if i reopen the app manually, i am back at the sign-in phase.” [source](https://twitter.com/2098621266370703360/status/2111608837409218572)
- Complaint, 2026-10-18, @AmpCode (X): “@ampcode @sqs mac app suddenly logged me out and it fails to log-in. i authenticate within the browser, then i click the oauth link, it opens amp but it doesn't log-in. was working like an hour ago” [source](https://twitter.com/1292843991197716483/status/2111662711549391069)
- Complaint, 2026-10-17, @AmpCode (X): “@ampcode i got logged out of the mac desktop app, and cannot log back in. webapp and ios works. can you help?” [source](https://twitter.com/237219814/status/2111464414759022664)

### Choosing models: 5th of 12 (customer love 0.559, n 77)

Ratings on models sit above the category norm, though praise and complaints come out about even. Offering the models users want ranks 7th of 11.

| Criterion | Reading | Customer love | 95% interval | n | Praise | Complaint |
|---|---|---|---|---|---|---|
| [Which models are offered on a plan and when](https://feedbackbench.com/coding-agents/criteria/models.catalog_access.md) | 7th of 11 | 0.551 | 0.524–0.578 | 37 | 21 | 16 |
| [Automatic model routing and fallback](https://feedbackbench.com/coding-agents/criteria/models.routing_auto.md) | Too few posts | 0.521 | 0.500–0.542 | 24 | 13 | 11 |
| [Reasoning effort setting and its defaults](https://feedbackbench.com/coding-agents/criteria/models.effort_control.md) | Too few posts | 0.489 | 0.477–0.501 | 16 | 4 | 12 |
| [Quality got worse or better over time](https://feedbackbench.com/coding-agents/criteria/models.quality_drift.md) | Too few posts | 0.514 | 0.502–0.529 | 7 | 5 | 2 |

Most recent posts:

- Praise, 2026-10-12, @AmpCode (X): “current @ampcode model usage :d opus 5.5 fucking slaps [link]” [source](https://twitter.com/1705384263867379712/status/2109512145046270017)
- Praise, 2026-10-12, @AmpCode (X): “i love that @ampcode now supports claude! one request, @thorstenball and @sqs is there any chance that project-based connections is a thing coming soon? (work vs play ai subs based on project settings).” [source](https://twitter.com/1405835361021042691/status/2109653765561504200)
- Praise, 2026-10-10, @AmpCode (X): “if i swear to codex native app for coding (most of our team does) why is @ampcode better? the astra model is so good to us.” [source](https://twitter.com/8243222/status/2108848349302714711)
- Complaint, 2026-10-18, @AmpCode (X): “@ampcode any chance we might get an option to tune a model’s thinking level? i am bummed that in raw model mode i can’t use sonnet medium. :/ unless y’all have just found it’s not worth the complexity or something?” [source](https://twitter.com/19434237/status/2111659214892659019)
- Complaint, 2026-10-12, @AmpCode (X): “seems like when one orb wants to spin up a claude code orb in @ampcode it doesn't get to pick the claude code model? i tried getting it to spin up opus 5.5 cc orbs but always get sonnet 5.5 cc orbs something there gentlemen?” [source](https://twitter.com/89691524/status/2109449297846751664)
- Complaint, 2026-10-12, @AmpCode (X): “@justine_chang39 @ampcode the model picker getting ignored by sub-agents is so annoying. does the orb config actually pass the model flag through?” [source](https://twitter.com/1643908924817670145/status/2109762293127094639)

### Instructing and context: 4th of 11 (customer love 0.511, n 40)

Users rarely bring up instructions and context: 4.4% of what they rate, against 10.0% for other agents. When they do, praise and complaints are about even, and ratings sit at the category norm.

| Criterion | Reading | Customer love | 95% interval | n | Praise | Complaint |
|---|---|---|---|---|---|---|
| [Direct in-prompt instructions and caps are followed](https://feedbackbench.com/coding-agents/criteria/context.instruction_following.md) | Too few posts | 0.510 | 0.498–0.524 | 10 | 6 | 4 |
| [Context compaction keeps what matters, cheaply and quickly](https://feedbackbench.com/coding-agents/criteria/context.compaction.md) | Too few posts | 0.510 | 0.500–0.520 | 9 | 7 | 2 |
| [Output degrades as the context window fills](https://feedbackbench.com/coding-agents/criteria/context.long_context_decay.md) | Too few posts | 0.498 | 0.491–0.505 | 7 | 1 | 6 |
| [Memory and state carried across sessions](https://feedbackbench.com/coding-agents/criteria/context.session_memory.md) | Too few posts | 0.500 | 0.493–0.507 | 6 | 3 | 3 |
| [Finding the right files in the codebase](https://feedbackbench.com/coding-agents/criteria/context.codebase_retrieval.md) | Too few posts | 0.497 | 0.492–0.501 | 4 | 1 | 3 |
| [Persistent project rules files are read and obeyed](https://feedbackbench.com/coding-agents/criteria/context.instruction_files.md) | Too few posts | 0.503 | 0.500–0.507 | 3 | 3 | 0 |
| [Asks the user versus guessing](https://feedbackbench.com/coding-agents/criteria/context.clarifying_questions.md) | Too few posts | 0.497 | 0.492–0.500 | 2 | 0 | 2 |
| [Images, PDFs and file attachments as input](https://feedbackbench.com/coding-agents/criteria/context.attachments.md) | Too few posts | 0.499 | 0.497–0.500 | 1 | 0 | 1 |

Most recent posts:

- Praise, 2026-10-09, @AmpCode (X): “@voip_guru @mattpocockuk yeah, auto compaction on @ampcode feels solved for me. i don’t think about it anymore.” [source](https://twitter.com/19506770/status/2108591171832537454)
- Praise, 2026-09-24, @AmpCode (X): “@kentcdodds @bot come to think of it, all the agentic tools i use the most, @ampcode @bot and some hermes, all of them abstract compaction away, and i have continuous sessions with all 3, and no dumb zones” [source](https://twitter.com/33135576/status/2103020461017977219)
- Praise, 2026-09-17, @AmpCode (X): “my absolutely favorite new @ampcode feature: recaps! amp gives you a summary of what happened in the thread if you have been away for a while. please more features that help me make sense of these dozens and dozens of agents! [link]” [source](https://twitter.com/631332723/status/2100484614838263984)
- Complaint, 2026-10-17, @AmpCode (X): “@nicolaygerold @ampcode i think 6.1 sol is treating the browser use skill as his almighty god and i cannot do anything to make him not follow it 🤣” [source](https://twitter.com/1705384263867379712/status/2111427919348482489)
- Complaint, 2026-10-02, @AmpCode (X): “@thorstenball @ampcode and to be clear, it isn’t so much a question of durability or being afraid i’ll lose it entirely. it’s more a question of durability in effectiveness in a single thread without having to be explicitly reminded to follow a specific playbook every few hours.” [source](https://twitter.com/7566122/status/2106000114821247188)
- Complaint, 2026-10-01, @AmpCode (X): “@thorstenball is there a way to keep some aspect of the context from compacting? i keep having issues where after long enough time, my orchestrator stops remembering the opening rules/playbook i give it and @ampcode starts to play fast and loose with me preferred processes.” [source](https://twitter.com/7566122/status/2105804263280840763)

### Doing the work: 2nd of 14 (customer love 0.621, n 238)

This area draws the most praise. Users say Amp handles their kind of task, and it ranks 1st of 10 on subagents working well. Long unattended runs also draw far more praise than complaints.

| Criterion | Reading | Customer love | 95% interval | n | Praise | Complaint |
|---|---|---|---|---|---|---|
| [Can do the user's kind of task](https://feedbackbench.com/coding-agents/criteria/work.capability.md) | 5th of 13 | 0.560 | 0.536–0.583 | 104 | 89 | 15 |
| [Subagents, parallel agents and orchestrators](https://feedbackbench.com/coding-agents/criteria/work.multi_agent_orchestration.md) | 1st of 10 | 0.556 | 0.533–0.578 | 87 | 76 | 11 |
| [Long unattended runs and goal/loop mode](https://feedbackbench.com/coding-agents/criteria/work.long_running_autonomy.md) | 4th of 8 | 0.509 | 0.492–0.525 | 36 | 30 | 6 |
| [Git commits, branches and sync](https://feedbackbench.com/coding-agents/criteria/work.git_workflow.md) | Too few posts | 0.491 | 0.479–0.502 | 15 | 2 | 13 |
| [Computer use and browser control](https://feedbackbench.com/coding-agents/criteria/work.computer_browser_use.md) | Too few posts | 0.505 | 0.496–0.515 | 13 | 8 | 5 |
| [Diagnosing and fixing reported bugs](https://feedbackbench.com/coding-agents/criteria/work.bug_diagnosis.md) | Too few posts | 0.501 | 0.491–0.510 | 9 | 7 | 2 |
| [Length and clarity of replies, summaries and comments](https://feedbackbench.com/coding-agents/criteria/work.response_verbosity.md) | Too few posts | 0.512 | 0.501–0.526 | 7 | 5 | 2 |
| [Stops mid-task or answers instead of acting](https://feedbackbench.com/coding-agents/criteria/work.premature_stop.md) | Too few posts | 0.496 | 0.492–0.499 | 6 | 0 | 6 |
| [Tool approval prompts and autonomy modes](https://feedbackbench.com/coding-agents/criteria/work.permission_prompts.md) | Too few posts | 0.509 | 0.496–0.523 | 5 | 3 | 2 |
| [Frontend and visual UI output](https://feedbackbench.com/coding-agents/criteria/work.frontend_ui.md) | Too few posts | 0.501 | 0.496–0.509 | 2 | 1 | 1 |
| [Does unrequested work or over-engineers](https://feedbackbench.com/coding-agents/criteria/work.scope_overreach.md) | Too few posts | 0.499 | 0.497–0.500 | 2 | 0 | 2 |
| [Plan-before-edit mode](https://feedbackbench.com/coding-agents/criteria/work.plan_mode.md) | Too few posts | 0.500 | 0.494–0.505 | 2 | 1 | 1 |
| [Breaks existing code or reintroduces bugs](https://feedbackbench.com/coding-agents/criteria/work.regressions_introduced.md) | Too few posts | 0.499 | 0.498–0.500 | 1 | 0 | 1 |
| [Risky or irreversible actions without confirmation](https://feedbackbench.com/coding-agents/criteria/work.destructive_actions.md) | Too few posts | 0.499 | 0.496–0.500 | 1 | 0 | 1 |
| [Caves to or argues with the user's judgement](https://feedbackbench.com/coding-agents/criteria/work.sycophancy_pushback.md) | Too few posts | 0.499 | 0.498–0.500 | 1 | 0 | 1 |
| [Spins, loops or gets stuck without progress](https://feedbackbench.com/coding-agents/criteria/work.stuck_loops.md) | Too few posts | 0.500 | 0.500–0.500 | 0 | 0 | 0 |
| [Games checks instead of fixing the problem](https://feedbackbench.com/coding-agents/criteria/work.reward_hacking.md) | Too few posts | 0.500 | 0.500–0.500 | 0 | 0 | 0 |
| [Safety filters block legitimate coding tasks](https://feedbackbench.com/coding-agents/criteria/work.safety_refusals.md) | Too few posts | 0.500 | 0.500–0.500 | 0 | 0 | 0 |

Most recent posts:

- Praise, 2026-10-18, @AmpCode (X): “@marccampbell the intergration @ampcode added recently is amazing. it watches for ci and reviews and automatically takes actions to get move the pr to a ready state.” [source](https://twitter.com/5444392/status/2111641515269501385)
- Praise, 2026-10-16, @AmpCode (X): “@ampcode blew my mind a little by _nesting_ a thread. agent did a series of tasks then asked how to continue. i typed "we'll start with #80 but in a new orb to pick up the merged changes". "i'll start a new orb thread…" then gave it relevant details from the old one. [link]” [source](https://twitter.com/9632112/status/2110885232081768675)
- Praise, 2026-10-16, @AmpCode (X): “my fiancé and i always forgot how to assemble the balloon arch lol. so, i spun up an orb in @ampcode and it showed step by step config, the screws we need and where to go. needed this for my sons 1st birthday party this saturday lol. [link]” [source](https://twitter.com/238651245/status/2111088770154205521)
- Complaint, 2026-10-10, @AmpCode (X): “@adamwathan @ampcode i tried it and really wanted to like it (i gave it 5 minutes, etc) but it seemed way more cumbersome than working with git worktrees locally.” [source](https://twitter.com/1264377739890561026/status/2108964241885864288)
- Complaint, 2026-10-09, @AmpCode (X): “starting to believe this is @ampcode's doing ....? i don't get this behavior from openai models in any other coding harnesses. eg: you paste in an error log and the model just responds with a vague possible causes rather than reading the source code and finding the bug. [link]” [source](https://twitter.com/1437462200/status/2108434439240090051)
- Complaint, 2026-10-02, @AmpCode (X): “@jkudish @thorstenball @ampcode just feels weird that the general orchestration plan slowly fades over the course of several hours. suddenly realizing that it isn't even delegating to child threads at all starts to just do everything itself.” [source](https://twitter.com/7566122/status/2105830588813087149)

### Checking and finishing: 4th of 9 (customer love 0.509, n 34)

Users seldom discuss how Amp checks its own work. Praise and complaints are close to even, and ratings sit at the category norm.

| Criterion | Reading | Customer love | 95% interval | n | Praise | Complaint |
|---|---|---|---|---|---|---|
| [Reviewing and approving the agent's changes](https://feedbackbench.com/coding-agents/criteria/verify.change_review_ui.md) | Too few posts | 0.505 | 0.490–0.521 | 20 | 10 | 10 |
| [Builds, tests or runs its own changes](https://feedbackbench.com/coding-agents/criteria/verify.self_testing.md) | Too few posts | 0.511 | 0.501–0.521 | 10 | 8 | 2 |
| [Agent-performed code review finds real issues](https://feedbackbench.com/coding-agents/criteria/verify.agent_code_review.md) | Too few posts | 0.500 | 0.490–0.507 | 5 | 4 | 1 |
| [Claims work is done or fixed when it is not](https://feedbackbench.com/coding-agents/criteria/verify.false_completion.md) | Too few posts | 0.498 | 0.496–0.500 | 2 | 0 | 2 |

Most recent posts:

- Praise, 2026-10-11, @AmpCode (X): “more bullish on @ampcode by the minute. it's supplying before/after screenshots, and inline interactive windows allowing you to check it's work right within the chat. here i am testing out a change to a tui demo app inline. [link]” [source](https://twitter.com/2016244386355433472/status/2109325036930900259)
- Praise, 2026-09-30, @AmpCode (X): “@adamwathan @ampcode threads as the new pr is how my team already reviews agent work, amp's visibility beat private chat dumps for us” [source](https://twitter.com/2055860494146170880/status/2105285491537092878)
- Praise, 2026-09-11, @AmpCode (X): “`amp sync` is yet another awesome addition from @ampcode - lets you (temporarily) sync changes from a thread to your machine and deletes them when you exit. great for review in your preferred diff tool.” [source](https://twitter.com/1491081/status/2098282433674092676)
- Complaint, 2026-10-09, @AmpCode (X): “@sqs @ampcode actually one thing - i think currently extra cloned projects do not show up in the changes panel, that becomes an issue.” [source](https://twitter.com/44927319/status/2108567641308975505)
- Complaint, 2026-10-06, @AmpCode (X): “@sqs @sarve___tanvesh @ampcode one thing i use the github pr review ui for is marking file diffs as viewed, which is not easily done in amp” [source](https://twitter.com/101315185/status/2107302886866624662)
- Complaint, 2026-09-29, @AmpCode (X): “@adamwathan @ampcode a thread isn’t a pr if i have to scroll through the whole agent run just to find the code changes worth reviewing” [source](https://twitter.com/1883915246084968455/status/2105010509091152053)

### Interface and sessions: 2nd of 14 (customer love 0.595, n 313)

Users like working from a phone on a cloud machine. Cloud sessions rank 1st of 5, and mobile and remote control also leads the category. The weak spot is display and settings, where users find the layout and controls confusing.

| Criterion | Reading | Customer love | 95% interval | n | Praise | Complaint |
|---|---|---|---|---|---|---|
| [How the interface shows work, and what the user can configure](https://feedbackbench.com/coding-agents/criteria/ui.display_settings.md) | 8th of 12 | 0.480 | 0.455–0.507 | 140 | 43 | 97 |
| [Cloud and remote sandbox execution](https://feedbackbench.com/coding-agents/criteria/surfaces.cloud_sessions.md) | 1st of 5 | 0.590 | 0.563–0.618 | 118 | 96 | 22 |
| [Mobile, remote-control and voice access](https://feedbackbench.com/coding-agents/criteria/surfaces.remote_mobile.md) | 1st of 7 | 0.559 | 0.538–0.580 | 64 | 54 | 10 |
| [Saving, switching, resuming and rewinding sessions](https://feedbackbench.com/coding-agents/criteria/ui.session_history.md) | Too few posts | 0.512 | 0.492–0.533 | 29 | 10 | 19 |
| [Stopping and steering a running agent](https://feedbackbench.com/coding-agents/criteria/ui.interrupt_steer.md) | Too few posts | 0.503 | 0.498–0.512 | 4 | 2 | 2 |

Most recent posts:

- Praise, 2026-10-18, @AmpCode (X): “i started using @ampcode on ios to trigger some small things on the go, or discuss some idea, and it’s such a wonderful experience i’d recommend anyone to try” [source](https://twitter.com/16185220/status/2111887423152222671)
- Praise, 2026-10-17, @AmpCode (X): “@d4m1n my guess is that i already do around 20% of my product building from my iphone, @ampcode is the killer iphone duo app, like evernote was for the first ipad” [source](https://twitter.com/14405744/status/2111430017448058963)
- Praise, 2026-10-16, @AmpCode (X): “@ampcode much much much cleaner look! kudos to who did this!🫡 [link]” [source](https://twitter.com/838785842/status/2110910193341223264)
- Complaint, 2026-10-15, @AmpCode (X): “@ampcode @thorstenball my amp wish list for christmas this year: linux desktop client and android native client. i'm not made for that pwa life. 🫡” [source](https://twitter.com/56608163/status/2110548452375658879)
- Complaint, 2026-10-15, @AmpCode (X): “@rockatanescu @sqs @ampcode oh that's where you change it, i was expecting a dropdown on the model router page... so i couldn't figure out where else it could be. a weird dial, right :) haha” [source](https://twitter.com/151126971/status/2110776770345513421)
- Complaint, 2026-10-14, @AmpCode (X): “@sqs @jbrooksuk @ampcode also auto-spelling corrections? or at least a toggle to turn that off? i have tons of domain specific words or random internal acronyms that get auto-corrected into something i need to recorrect” [source](https://twitter.com/2000470921329709056/status/2110398803522064750)

### Reliability and speed: 6th of 13 (customer love 0.521, n 103)

Users complain about this area more than they praise it. Complaints concern service errors and client crashes. Even so, Amp ranks 1st of 6 on few outages and errors.

| Criterion | Reading | Customer love | 95% interval | n | Praise | Complaint |
|---|---|---|---|---|---|---|
| [Outages, server errors and capacity or rate errors](https://feedbackbench.com/coding-agents/criteria/rel.service_errors.md) | 1st of 6 | 0.514 | 0.486–0.542 | 41 | 5 | 36 |
| [Client crashes, freezes and failed tool execution](https://feedbackbench.com/coding-agents/criteria/rel.client_failures.md) | 5th of 11 | 0.511 | 0.477–0.549 | 35 | 3 | 32 |
| [Latency, throughput and fast mode](https://feedbackbench.com/coding-agents/criteria/rel.response_speed.md) | Too few posts | 0.513 | 0.493–0.535 | 25 | 9 | 16 |
| [Updates break working setups](https://feedbackbench.com/coding-agents/criteria/rel.update_breakage.md) | Too few posts | 0.515 | 0.497–0.537 | 8 | 3 | 5 |

Most recent posts:

- Praise, 2026-10-15, @AmpCode (X): “one observation from using @ampcode : opus seems to churn through coding tasks noticeably faster than gpt sol. been surprised by how quickly it's putting up prs. curious if the amp team has noticed similar differences in coding throughput between the two models.” [source](https://twitter.com/2016294660658941952/status/2110578795505680872)
- Praise, 2026-10-08, @AmpCode (X): “@taylorotwell @ampcode i was also pretty shocked at how fast it ran everything. idk how they’re doing that” [source](https://twitter.com/19832122/status/2108097691523690697)
- Praise, 2026-09-30, @AmpCode (X): “@ruchernchong @ampcode it's way more responsive. no lag, no blur, just better to use!” [source](https://twitter.com/9111552/status/2105262479638958295)
- Complaint, 2026-10-18, @AmpCode (X): “@ampcode please, threads are failing to load and now the desktop app is dead too [link]” [source](https://twitter.com/1138798227111776257/status/2111753490041061801)
- Complaint, 2026-10-16, @AmpCode (X): “@ampcode had some slow experience accessing things from apac, i just feel jealous for anyone living in the us with minimum latency when orbin' lol” [source](https://twitter.com/2022713436736753664/status/2111214216782987644)
- Complaint, 2026-10-15, @AmpCode (X): “these default options are just too small @ampcode . i am running out of memory all the time. [link]” [source](https://twitter.com/2792677436/status/2110768587015147944)

### Account and support: 1st of 11 (customer love 0.624, n 50)

Users praise the support team. Amp ranks 1st of 8 on helpful support, and praise outnumbers complaints.

| Criterion | Reading | Customer love | 95% interval | n | Praise | Complaint |
|---|---|---|---|---|---|---|
| [Support, refunds and issue handling](https://feedbackbench.com/coding-agents/criteria/account.support.md) | 1st of 8 | 0.612 | 0.575–0.652 | 38 | 29 | 9 |
| [Account bans and access restrictions](https://feedbackbench.com/coding-agents/criteria/account.bans_restrictions.md) | Too few posts | 0.505 | 0.491–0.530 | 8 | 1 | 7 |
| [Wrong charges, failed payments and plan provisioning](https://feedbackbench.com/coding-agents/criteria/account.billing_errors.md) | Too few posts | 0.497 | 0.495–0.499 | 4 | 0 | 4 |
| [Data retention, training use and deployment isolation](https://feedbackbench.com/coding-agents/criteria/account.data_privacy.md) | Too few posts | 0.501 | 0.495–0.511 | 4 | 1 | 3 |

Most recent posts:

- Praise, 2026-10-15, @AmpCode (X): “@tobimori @steveruizok @chatgpt @ampcode oh sure. yeah, no tool is perfect. i'm sending minor bug reports / feature requests to them all the time. but amp is good, thoughtful, forward-thinking, they listen, and they ship!” [source](https://twitter.com/1796637816589529088/status/2110778863471177799)
- Praise, 2026-10-13, @AmpCode (X): “reported a bug in @ampcode, less than 30 minutes later i get a personal(!) email that it's solved and will be rolled out later today. wow! this is the bar now! kudos @nicolaygerold” [source](https://twitter.com/2385015511/status/2109987646034325646)
- Praise, 2026-10-08, @AmpCode (X): “@ampcode. this is a sweet way to acknowledge bug reporting and its fixed too [link]” [source](https://twitter.com/38651218/status/2108034167082811845)
- Complaint, 2026-10-13, @AmpCode (X): “@ampcode does [email] work? i send requests there, but i don't get confirmation that it was received nor hear back from anyone.” [source](https://twitter.com/19506770/status/2110018658282160213)
- Complaint, 2026-09-29, @AmpCode (X): “@sqs @romainlanz @ampcode i get why it works that way, but honestly still feels weird from a privacy standpoint even if no one's looking, knowing they could makes me feel watched/spied. makes me think twice about using amp more, even though i'd like to but maybe it's just me 🤓” [source](https://twitter.com/697207221097275392/status/2105063286995558713)
- Complaint, 2026-09-23, @AmpCode (X): “@sqs @ampcode happened to us as well. we gave up on support and about to setup new accounts. seems like very common issue.” [source](https://twitter.com/1089515309122367489/status/2102754710373691815)