What engineering leaders and their teams report about Amp, with every finding tied to its evidence.
Read the findingsPractitioner sentiment, all practitioners
+69 Strong positive from 56 sources that take a side
How this figure is madeFrom 51 authors in 53 discussions. Use a task pilot to measure accepted work and review time.
Jun 2025 to Oct 2026.
77 sources from practitioners, by month and sentiment.
Across the 56 sources that give an opinion the read is strongly positive, led by praise for model routing, result quality, and orbs. Cost is the most discussed concern. Most engineers who compared agents keep Amp as their main agent, and 9 use another.
Positive, mixed and negative sources
The score counts them as 38 positive, 10 mixed and 3 negative votes, one per author. See How the numbers work.
See the evidenceBelow: 6 praise themes, 1 concern, 1 close to agreement and 2 disagreements.
Counted for All practitioners, from 77 sources. Change group
Every capability practitioners describe positively, ranked by how many sources say so. Open one to read who said it.
Counted for All practitioners.
Positive, 15
Positive
microdrum, Hacker News commenter, self-described
Comment on: Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-AstraHacker News comment
Says Amp has led coding agents for almost a year.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Positive
ahmadyan, Hacker News commenter
Comment on: DeepClaude - Claude Code agent loop with DeepSeek V4 ProHacker News comment
Finds the same Opus model works better in third-party harnesses such as Amp and Factory Droid than in Claude Code.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
shell0x, Hacker News commenter, self-described
Comment on: How I write software with LLMsHacker News comment
Has used Codex and Claude Max but keeps returning to Amp, which routes models by mode and often gets things right on the first attempt.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Mixed
Web developer
AMP vs Claude Code (CLI): Comparing of Agentic Coding ToolsPersonal blog post, cesarcouto.com
Ran Amp and Claude Code with the same models, repository, and AGENTS.md; Amp was faster, cleaner, and needed fewer fixes, but only its free tier made it affordable.
Practitioner account, self-described; structured test; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Positive
truekonrads, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Loves Amp's results and its opinionated design, including prompts to hand off context, and not being tied to one model.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
pjm331, Hacker News commenter, self-described
Comment on: Agent design is still hardHacker News comment
Says Amp started a bit rough but is now really good.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
derencius, Hacker News commenter, self-described
Comment on: Getting good results from Claude CodeHacker News comment
Compares terminal agents and rates Amp very good and precise with its changes.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Positive
sergiotapia, Hacker News commenter, self-described
Comment on: 6 weeks of Claude CodeHacker News comment
Has used Amp heavily and says it always hits the target.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Independent developer and contributor to the Air framework
Amp Code Deep Dive and TalkPersonal blog post, Elite AI Assisted Coding
Used Amp for a real open-source documentation task that was merged; it fixed a database bug from logs on the first try for US$4.10, though drafts needed trimming.
Practitioner account, self-described; structured test; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
mrklol, Hacker News commenter, self-described
Comment on: Clarifying our pricingHacker News comment
Tells others to just switch to Amp, saying its agent is superior in every way and its tooling achieves even more although it uses the same models.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Positive
Kieran Klaassen, general manager of Cora at Every
Transcript: 'How Two Engineers Ship Like a Team of 15 With AI Agents'Podcast episode transcript, AI & I (Every)
Ranking AI coding agents on the Every podcast, Klaassen puts Amp in the S tier under Claude Code, saying it is very good at just getting work done, with pretty good ergonomics and good tools already.
"Yes. It's very good at just getting work done. The ergonomics are pretty good, good tools already."
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Positive
Andrzej Krzywda, CEO of Arkency
Post on an agent recording a bug video in an orbX post
Had the Amp agent find a bug and then record a video of it inside an orb, a workflow he says orbs made frictionless.
"Having the agent finding the bug, then record a video of a bug, this is a new addition to my toolbox."
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Mixed
sergiotapia, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Amp was their primary workhorse until they moved to cheaper models in the Claude CLI and then Opus 4.5 directly, spending about US$5 a day on Amp.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
hnidiots3, Hacker News commenter, self-described
Comment on: Claude Code on the webHacker News comment
Says having Amp check everything with the Oracle, a second model, gives better results than Codex on backend work.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Mixed
Developer
Amp - Vibe Code With Me - Ep 016Video, YouTube
Calls Amp the best coding agent used so far and by far the most expensive, at US$15 in two hours. It was the first test where all the tasks got done, yet the author finds the return hard to justify against Claude Code Max at US$100 a month.
"…could be interesting. But I still think I think I still think it's hard to justify this cost compared to like just giving everybody Claude Max, for example, you know, the 5x or even the 10X. Again, I spent $15 in two hours."
Practitioner account, self-described; structured test; counts toward Delivered outcomes, Reliability, Cost predictability and limits, Adoption and upkeep
Mixed or negative, 1
Mixed
Independent developer maintaining open-source side projects
AI coding agent one-shot challenge with Amp, Claude Code and GeminiPersonal blog post, Land of Unknwon (unknwon.io)
Gave Amp, Claude Code, and Gemini CLI the same one-shot build; Amp matched the template structure best but left a broken app and stopped mid-task.
Practitioner account; structured test; counts toward Delivered outcomes, Reliability
Counted for All practitioners.
Mixed
Software engineer
Amp vs Claude Code: Is It Worth Switching in 2026?Personal blog post, alexdunlop.com
Ran Amp and Claude Code on the same Python task: Amp's model routing and visible subagents stood out, it felt expensive, and both failed on ruff.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
jauntywundrkind, Hacker News commenter, self-described
Comment on: Amp Free Is FullHacker News comment
Calls Amp an amazing product they love recommending, praising handoff, threads and sharing, and the Oracle. Says people sign up right away once told about the free, ad-supported plan.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Positive
incoming1211, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Compares most terminal agents and says Amp costs more but gives the best results, crediting the Librarian and the Oracle.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Co-founder of Octoco
Getting Started with AmpPersonal blog post, G-J's Substack
A co-founder's starter guide for Octoco's senior engineers: keep threads short, plan with the Oracle, hand off phases, and expect cost to grow with long threads.
Practitioner account, self-described; direct use; counts toward Reliability, Operational control, Cost predictability and limits
Positive
hnidiots3, Hacker News commenter, self-described
Comment on: Claude Code on the webHacker News comment
Says having Amp check everything with the Oracle, a second model, gives better results than Codex on backend work.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Positive
Creator of Ghostty
Vibing a Non-Trivial Ghostty FeaturePersonal blog post, mitchellh.com
Built a Ghostty feature with Amp over 16 sessions totalling US$15.98 in token spend, consulting the Oracle for all planning. Guided the agent through cleanup sessions with minor manual changes and merged after a final manual review.
Practitioner account; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Mixed
pjm331, Hacker News commenter, self-described
Comment on: Jules, our asynchronous coding agentHacker News comment
Finds Amp good but short of Claude Code's polish and features, and reaches for it for code review because of the Oracle.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
mark_undoio, CTO of Undo, Hacker News commenter
Comment on: Claude Opus 4.1Hacker News comment
Found that asking Amp to have the Oracle check the main model's debugging work helped, and the main model did better once prompted that way.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
d_sc, Hacker News commenter, self-described
Comment on: Cognition (Devin AI) to Acquire WindsurfHacker News comment
Rates Amp very good, highlighting the Oracle that uses o3 for reviews and a team leaning into the agentic approach.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Positive
usef-, Hacker News commenter, self-described
Comment on: Claude Sonnet 5Hacker News comment
Recalls that Amp picked the best model for each task and that its CLI steered users toward good practice, such as replacing a context clear with a handoff.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
shell0x, Hacker News commenter, self-described
Comment on: How I write software with LLMsHacker News comment
Has used Codex and Claude Max but keeps returning to Amp, which routes models by mode and often gets things right on the first attempt.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
Developer
Should You Use Ampcode for Production Code? One Month InPersonal blog post, IntelligentTools
After a month with both Cursor and Amp, uses Amp for refactorings and architectural changes that touch many files, and Deep mode almost exclusively for production work. Runs Amp on a dedicated branch and reviews every commit before merging to main.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Reliability, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
truekonrads, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Loves Amp's results and its opinionated design, including prompts to hand off context, and not being tied to one model.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Independent developer
Amp: Notes from reading the Amp manualPersonal blog post, hamel.dev
Names threads their favorite Amp feature, finds model abstraction works despite early doubts, and pays per token because the time saved is worth it.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Counted for All practitioners.
Positive, 14
Positive
cavoirom, Hacker News commenter, self-described
Comment on: Plan mode is deadHacker News comment
Has used Amp since November 2025 and says it needs no plan mode, only conversation.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
usef-, Hacker News commenter, self-described
Comment on: Claude Sonnet 5Hacker News comment
Recalls that Amp picked the best model for each task and that its CLI steered users toward good practice, such as replacing a context clear with a handoff.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
4b11b4, Hacker News commenter, self-described
Comment on: Claude Code is being dumbed down?Hacker News comment
Says Amp feels much better than Claude Code and likes that a message reaches the agent as soon as they send it.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Independent developer
I keep coming back to Amp CodePersonal blog post, Jibran's Perspective (asadjb.com)
Compared Amp, Claude Code, and Codex on the same tasks and enjoyed Amp most; cost pushed them to Claude Max, but they plan to return to Amp.
Practitioner account; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
tbliu, Hacker News commenter, self-described
Comment on: Anthropic blocks third-party use of Claude Code subscriptionsHacker News comment
Uses Amp's free tier with Opus 4.5 heavily, calls it a very nice CLI, and finds the ads fairly unobtrusive.
Practitioner account, pseudonymous; direct use; counts toward Operational control, Cost predictability and limits
Positive
the_mitsuhiko, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Thinks Amp is great but pricey, and says it feels like a product its own builders use.
Practitioner account, pseudonymous; opinion; not counted, takes no position
Positive
Developer and streamer
Can AMP Code Beat Claude Code? | Feature Build Day 3Video (livestream), YouTube
Tried Amp as a plugin in Cursor to see if it could beat a Claude Code workflow while adding a photo upload feature live. Had Amp turn existing Claude.md files and Cursor rules into AGENTS.md files and calls the agent really, really smooth and really easy.
"This is nice. And now we have a pages, so we have go to app page. I'm just kind of reviewing the code as it's flying in. So client-side validation type, great. Awesome. I think so it just feels smooth I a…"
Practitioner account, self-described; direct use; counts toward Adoption and upkeep
Positive
Gys, Hacker News commenter, self-described
Comment on: Ask HN: Why do Cursor, Windsurf and Claude Code dominate the conversation?Hacker News comment
Uses the Amp VS Code plugin and is very happy with it, finding it easier than Claude Code and Cursor, which they tried earlier.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Kieran Klaassen, general manager of Cora at Every
Transcript: 'How Two Engineers Ship Like a Team of 15 With AI Agents'Podcast episode transcript, AI & I (Every)
Ranking AI coding agents on the Every podcast, Klaassen puts Amp in the S tier under Claude Code, saying it is very good at just getting work done, with pretty good ergonomics and good tools already.
"Yes. It's very good at just getting work done. The ergonomics are pretty good, good tools already."
Practitioner account, self-described; direct use; counts toward Delivered outcomes
Positive
vladstudio, Hacker News commenter
Comment on: Cursor 1.0Hacker News comment
Says Amp took model and prompt configuration off their hands, and that they enjoy AI-assisted coding more because it just works.
Practitioner account, pseudonymous; direct use; counts toward Adoption and upkeep
Positive
Andrei Maxim, applied AI engineer and former Ruby developer
The State of Agents in the CloudX post (X article)
After two weeks on Amp, set up orbs, Claude Code cloud, and Codex cloud side by side. Amp generated its own orb setup scripts and orbs just worked, which he rates well above the other two cloud agents.
"All in all, Amp with Orbs feels significantly above the current state of agents in the cloud, much like the jump from Opus 4.1 to 4.5 from late last year."
Practitioner account, pseudonymous; direct use; counts toward Adoption and upkeep
Positive
Mike Hostetler, founder and Elixir agent framework builder
Amp Got Cloud Coding Sandboxes RightX post (X article)
Pointed an orb at his ReqLLM repository: Amp prepared the environment with a skill, made the change, and pushed it to GitHub with permissions in place. His complaints were a large default orb size and a confusing project screen.
"The agent made the requested change, committed the code, and pushed it back to GitHub with the permissions already in place."
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Mixed
Web developer
AMP vs Claude Code (CLI): Comparing of Agentic Coding ToolsPersonal blog post, cesarcouto.com
Ran Amp and Claude Code with the same models, repository, and AGENTS.md; Amp was faster, cleaner, and needed fewer fixes, but only its free tier made it affordable.
Practitioner account, self-described; structured test; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Positive
WEBdoze, YouTube channel
I Tried the Ad-Supported AI Coding Agent (Amp Free)Video, WEBdoze (YouTube)
Tested Amp Free in the CLI and VS Code extension on a solar panel website, found the free result decent and functional, likes the UI and finds the in-tool ads pretty good. Guesses complex work will still need the paid smart option.
"…this is working. For complex things, I guess you will going to need still the smart option to be enabled. But yeah, you need to add money. I'm curious to see if this uh plan it will work also the free one and the one…"
Practitioner account, self-described; direct use; counts toward Operational control, Cost predictability and limits
Mixed or negative, 3
Mixed
Company founder, co-host of Mostly Technical
147: It's Orbin' TimePodcast episode, Mostly Technical (podcast)
Moved their company work to orbs to keep agents off their laptop and share running threads, but finds the weakest part is Amp's harness, and the ship button is loose.
"…button, but the ship button just does just injects a prompt. So it can do literally anything you want. And so it says yeah, it just says pull down uh fix any merge issues, you know, push it back up. Uh you don't need to…"
Practitioner account, self-described; opinion; not counted, takes no position
Mixed
pjm331, Hacker News commenter, self-described
Comment on: Jules, our asynchronous coding agentHacker News comment
Finds Amp good but short of Claude Code's polish and features, and reaches for it for code review because of the Oracle.
Practitioner account, pseudonymous; direct use; counts toward Operational control
Mixed
Developer and YouTube creator
Coding with Claude Opus 4.5 For Free! (AmpCode)Video, YouTube
Uses the Amp extension's free daily US$10 for a few Opus 4.5 requests beside their main tool, and dislikes the CLI.
"…run the code. It has two modes. Smart that uses OPUS 4.5, GBT 5.2 and also RAS which uses smaller models like Hiku 4.5. Now, this is super powerful because $10 of free usage every single day means you could use Oppus 4.5 and JP…"
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Counted for All practitioners.
Mixed
navs, Hacker News commenter
Comment on: How Compaction Works in PiHacker News comment
Says Amp had a handoff feature for a while that they found genuinely useful and, anecdotally, felt worked better than compaction, until Amp removed it.
Practitioner account, pseudonymous; direct use; counts toward Adoption and upkeep
Positive
SatvikBeri, Hacker News commenter, self-described
Comment on: Claude Code's compaction discards data that's still on diskHacker News comment
Uses Amp over Claude Code for its thread tree and handoff, which keep each session focused and context low; finds Claude Code's compaction worse.
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Positive
jauntywundrkind, Hacker News commenter, self-described
Comment on: Amp Free Is FullHacker News comment
Calls Amp an amazing product they love recommending, praising handoff, threads and sharing, and the Oracle. Says people sign up right away once told about the free, ad-supported plan.
Practitioner account, pseudonymous; opinion; not counted, repeats a stronger account
Positive
truekonrads, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Loves Amp's results and its opinionated design, including prompts to hand off context, and not being tied to one model.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
MorehouseJ09, Founder and CEO of Nuon, Hacker News commenter
Comment on: You should still be writing code from your editorHacker News comment
Is most bullish on Amp because it can be driven from different contexts and its threads beat anything else they have seen.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Independent developer
Amp Coding Agent: Features That Make It Stand OutPersonal blog post, peterwarnock.com
Says Amp has consistently impressed them, singling out handoff as a better version of Claude Code's compaction and scoped instructions for complex projects.
Practitioner account; direct use; counts toward Delivered outcomes
Positive
Independent developer
Amp: Notes from reading the Amp manualPersonal blog post, hamel.dev
Names threads their favorite Amp feature, finds model abstraction works despite early doubts, and pays per token because the time saved is worth it.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Mitchell Hashimoto, creator of Ghostty and co-founder of HashiCorp
Post on shared Amp threads in pull requestsX post
Says sharing a whole session globally is the favorite part of Amp, that pull requests with Amp threads attached make Hashimoto very, very happy as a maintainer, and links one from a bug fix that morning.
"PRs with Amp threads attached make me very, very happy as a maintainer."
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
usef-, Hacker News commenter, self-described
Comment on: Claude Sonnet 5Hacker News comment
Recalls that Amp picked the best model for each task and that its CLI steered users toward good practice, such as replacing a context clear with a handoff.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
Software engineer
Amp vs Claude Code: Is It Worth Switching in 2026?Personal blog post, alexdunlop.com
Ran Amp and Claude Code on the same Python task: Amp's model routing and visible subagents stood out, it felt expensive, and both failed on ruff.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Co-founder of Octoco
Getting Started with AmpPersonal blog post, G-J's Substack
A co-founder's starter guide for Octoco's senior engineers: keep threads short, plan with the Oracle, hand off phases, and expect cost to grow with long threads.
Practitioner account, self-described; direct use; counts toward Reliability, Operational control, Cost predictability and limits
Positive
Creator of Ghostty
Vibing a Non-Trivial Ghostty FeaturePersonal blog post, mitchellh.com
Built a Ghostty feature with Amp over 16 sessions totalling US$15.98 in token spend, consulting the Oracle for all planning. Guided the agent through cleanup sessions with minor manual changes and merged after a final manual review.
Practitioner account; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Mixed
Independent developer maintaining open-source side projects
AI coding agent one-shot challenge with Amp, Claude Code and GeminiPersonal blog post, Land of Unknwon (unknwon.io)
Gave Amp, Claude Code, and Gemini CLI the same one-shot build; Amp matched the template structure best but left a broken app and stopped mid-task.
Practitioner account; structured test; counts toward Delivered outcomes, Reliability
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.Handoff left with the May 2026 rebuild. Amp now compacts a thread automatically at 90% of its context window, and threads can still read and reference other threads. Praise for handoff describes the earlier product. Amp, Rebuilt
Open them in the recordCounted for All practitioners.
Positive
Taylor Otwell, creator of Laravel
Post on moving Omega development to the cloudX post
Moved his dev work to the cloud, tried Claude and Amp, and kept Amp because it got his workspace scripted while Claude kept hitting snags. He works across 5 repositories and finds orbs less locked down than a sandbox.
"Stuck with Amp because it was able to help me get my workspace scripted and working well - Claude was close but I just kept hitting random snags."
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes, Operational control, Adoption and upkeep
Mixed
Company founder, co-host of Mostly Technical
147: It's Orbin' TimePodcast episode, Mostly Technical (podcast)
Moved their company work to orbs to keep agents off their laptop and share running threads, but finds the weakest part is Amp's harness, and the ship button is loose.
"…button, but the ship button just does just injects a prompt. So it can do literally anything you want. And so it says yeah, it just says pull down uh fix any merge issues, you know, push it back up. Uh you don't need to…"
Practitioner account, self-described; opinion; not counted, takes no position
Positive
josevalerio, Hacker News commenter, self-described
Comment on: OrbsHacker News comment
Finds orbs more consistent than Codex cloud, alongside Cursor's cloud agents.
Practitioner account, pseudonymous; opinion; not counted, takes no position
Positive
altschuler, Hacker News commenter, self-described
Comment on: OrbsHacker News comment
Finds orbs fantastic for portals, multiplayer, local syncing, and Puck, and calls Amp the best US$20 a month they spend on dev tools.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Joey Kudish, software developer with 20 years of experience
Post on moving project work to AmpX post
Is moving most of his project work to Amp orbs and writing custom plugins and infrastructure to manage them, keeping Codex for occasional local work and computer use.
"I think I'm ready to move most of my project work to @AmpCode. I'm even writing some custom plugins and custom infra around it to better manage my orbs the way i need them to work."
Practitioner account, pseudonymous; direct use; counts toward Operational control
Positive
JJ Englert, Enablement Lead at Tenex
Post on the team's use of Amp orbsX post
Says his team is loving Amp's orbs, one of the few tools they found that makes moving work off local machines and into the cloud easy.
"Our team is loving amp right now, specifically how their orbs work."
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Andrzej Krzywda, CEO of Arkency
Post on an agent recording a bug video in an orbX post
Had the Amp agent find a bug and then record a video of it inside an orb, a workflow he says orbs made frictionless.
"Having the agent finding the bug, then record a video of a bug, this is a new addition to my toolbox."
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes
Positive
Ravindra Barthwal, founder building dxcode
Post to the Amp team on orbsX post
Found Amp good but costly in the past. After 4 days on the US$20 plan he gets a lot of value, with one orb working for over 4 days and coordinating with other orbs.
"In just $20 I am getting ton of value. One of my orb is working for over 4 days now and co-coordinating with other orbs…"
Practitioner account, pseudonymous; direct use; counts toward Reliability
Positive
r5Khe, Hacker News commenter, self-described
Comment on: Launch HN: Hoplite (YC S26) - Effortlessly deploy cloud coding agentsHacker News comment
Has used Amp for a while, says Amp "seems to be doing something very similar" to the cloud agent tool being launched, and calls one thread per VM a solid model going forward.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Andrei Maxim, applied AI engineer and former Ruby developer
The State of Agents in the CloudX post (X article)
After two weeks on Amp, set up orbs, Claude Code cloud, and Codex cloud side by side. Amp generated its own orb setup scripts and orbs just worked, which he rates well above the other two cloud agents.
"All in all, Amp with Orbs feels significantly above the current state of agents in the cloud, much like the jump from Opus 4.1 to 4.5 from late last year."
Practitioner account, pseudonymous; direct use; counts toward Adoption and upkeep
Positive
Developer, formerly at Heptabase
AmpCode is coolPersonal blog post, blog.ianhsiao.me
Uses orbs and thread-to-thread messaging to move work between local and remote machines, and credits Amp for removing features as models improve.
Practitioner account, self-described; direct use; counts toward Adoption and upkeep
Positive
Mike Hostetler, founder and Elixir agent framework builder
Amp Got Cloud Coding Sandboxes RightX post (X article)
Pointed an orb at his ReqLLM repository: Amp prepared the environment with a skill, made the change, and pushed it to GitHub with permissions in place. His complaints were a large default orb size and a confusing project screen.
"The agent made the requested change, committed the code, and pushed it back to GitHub with the permissions already in place."
Practitioner account, pseudonymous; direct use; counts toward Delivered outcomes, Operational control, Adoption and upkeep
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.Orbs bill by the minute, from US$0.08 per hour for a1.tiny to US$2.13 per hour for a1.3xlarge. They pause 5 minutes after the agent stops or 20 minutes after the last interaction, whichever is later. Runners on your own machines cost nothing. Sizes & Costs
Open them in the recordCounted for All practitioners.
Positive, 3
Positive
Ravindra Barthwal, founder building dxcode
Post to the Amp team on orbsX post
Found Amp good but costly in the past. After 4 days on the US$20 plan he gets a lot of value, with one orb working for over 4 days and coordinating with other orbs.
"In just $20 I am getting ton of value. One of my orb is working for over 4 days now and co-coordinating with other orbs…"
Practitioner account, pseudonymous; direct use; counts toward Reliability
Positive
altschuler, Hacker News commenter, self-described
Comment on: OrbsHacker News comment
Finds orbs fantastic for portals, multiplayer, local syncing, and Puck, and calls Amp the best US$20 a month they spend on dev tools.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Independent developer
Amp: Notes from reading the Amp manualPersonal blog post, hamel.dev
Names threads their favorite Amp feature, finds model abstraction works despite early doubts, and pays per token because the time saved is worth it.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Mixed or negative, 6
Mixed
SatvikBeri, Hacker News commenter, self-described
Comment on: Elevated error rates on Opus 4.7Hacker News comment
Has used Amp a bit and finds it quite nice, but more expensive than Claude without a subsidized subscription, so they have little reason to switch.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Independent developer
I keep coming back to Amp CodePersonal blog post, Jibran's Perspective (asadjb.com)
Compared Amp, Claude Code, and Codex on the same tasks and enjoyed Amp most; cost pushed them to Claude Max, but they plan to return to Amp.
Practitioner account; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Mixed
Web developer
AMP vs Claude Code (CLI): Comparing of Agentic Coding ToolsPersonal blog post, cesarcouto.com
Ran Amp and Claude Code with the same models, repository, and AGENTS.md; Amp was faster, cleaner, and needed fewer fixes, but only its free tier made it affordable.
Practitioner account, self-described; structured test; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Mixed
sergiotapia, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Amp was their primary workhorse until they moved to cheaper models in the Claude CLI and then Opus 4.5 directly, spending about US$5 a day on Amp.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
lvl155, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Found Amp really good in summer 2025 but calls cost its biggest downside and questions what the price buys without its own model.
Practitioner account, pseudonymous; opinion; not counted, takes no position
Negative
embedding-shape, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Tried Amp a couple of times, spending US$5, then US$10, then US$20, and says it is so expensive they struggle to judge it fairly. Has ChatGPT Pro and is not stingy with paying for useful things, but Amp is past their limit.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Every concern practitioners raise, most raised first.
Counted for All practitioners.
Raised as a concern, 11
Mixed
Software engineer
Amp vs Claude Code: Is It Worth Switching in 2026?Personal blog post, alexdunlop.com
Ran Amp and Claude Code on the same Python task: Amp's model routing and visible subagents stood out, it felt expensive, and both failed on ruff.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Mixed
SatvikBeri, Hacker News commenter, self-described
Comment on: Elevated error rates on Opus 4.7Hacker News comment
Has used Amp a bit and finds it quite nice, but more expensive than Claude without a subsidized subscription, so they have little reason to switch.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
Web developer
AMP vs Claude Code (CLI): Comparing of Agentic Coding ToolsPersonal blog post, cesarcouto.com
Ran Amp and Claude Code with the same models, repository, and AGENTS.md; Amp was faster, cleaner, and needed fewer fixes, but only its free tier made it affordable.
Practitioner account, self-described; structured test; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Positive
the_mitsuhiko, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Thinks Amp is great but pricey, and says it feels like a product its own builders use.
Practitioner account, pseudonymous; opinion; not counted, takes no position
Mixed
sergiotapia, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Amp was their primary workhorse until they moved to cheaper models in the Claude CLI and then Opus 4.5 directly, spending about US$5 a day on Amp.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Mixed
lvl155, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Found Amp really good in summer 2025 but calls cost its biggest downside and questions what the price buys without its own model.
Practitioner account, pseudonymous; opinion; not counted, takes no position
Negative
embedding-shape, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Tried Amp a couple of times, spending US$5, then US$10, then US$20, and says it is so expensive they struggle to judge it fairly. Has ChatGPT Pro and is not stingy with paying for useful things, but Amp is past their limit.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Mixed
Developer
Amp - Vibe Code With Me - Ep 016Video, YouTube
Calls Amp the best coding agent used so far and by far the most expensive, at US$15 in two hours. It was the first test where all the tasks got done, yet the author finds the return hard to justify against Claude Code Max at US$100 a month.
"…could be interesting. But I still think I think I still think it's hard to justify this cost compared to like just giving everybody Claude Max, for example, you know, the 5x or even the 10X. Again, I spent $15 in two hours."
Practitioner account, self-described; structured test; counts toward Delivered outcomes, Reliability, Cost predictability and limits, Adoption and upkeep
Mixed
Adam Stacoviak, co-founder and podcast host, Changelog
Changelog #658: Adam Stacoviak on running agents side by sidePodcast episode with publisher transcript
Stacoviak runs Amp beside Claude Code with a product-manager agent reviewing both, uses Claude Code partly because Amp costs him more, and calls Amp's cost justified.
"...in this case one of them was Amp and the other one was Claude Code, because that one's cheaper for me. You're really expensive, by the way."
Positive
cavoirom, Hacker News commenter, self-described
Comment on: Our decision on Cursor following its acquisition by SpaceXHacker News comment
Recommends Amp although it is not popular on Hacker News, saying people who try it will not go back, though the price may be a bit expensive.
Practitioner account, pseudonymous; opinion; not counted, takes no position
Positive
Independent developer
I keep coming back to Amp CodePersonal blog post, Jibran's Perspective (asadjb.com)
Compared Amp, Claude Code, and Codex on the same tasks and enjoyed Amp most; cost pushed them to Claude Max, but they plan to return to Amp.
Practitioner account; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Discussed without concern, 15
Mixed
zingar, Hacker News commenter, self-described
Comment on: Claude Code: connect to a local model when your quota runs outHacker News comment
Switches from a Claude Pro plan to Amp Free about two thirds of the way through each day and finds the change in ergonomics jarring.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Mixed
Developer and YouTube creator
Coding with Claude Opus 4.5 For Free! (AmpCode)Video, YouTube
Uses the Amp extension's free daily US$10 for a few Opus 4.5 requests beside their main tool, and dislikes the CLI.
"…run the code. It has two modes. Smart that uses OPUS 4.5, GBT 5.2 and also RAS which uses smaller models like Hiku 4.5. Now, this is super powerful because $10 of free usage every single day means you could use Oppus 4.5 and JP…"
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Positive
theshrike79, Hacker News commenter, self-described
Comment on: Unauthenticated remote code execution in OpenCodeHacker News comment
Uses Claude Code and Codex and recommends Amp alongside them for small utility scripts and changes on the free tier.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
incoming1211, Hacker News commenter
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Says their employer pays for Amp and that each member of their team spends more than US$1,000 a month on it.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
WEBdoze, YouTube channel
I Tried the Ad-Supported AI Coding Agent (Amp Free)Video, WEBdoze (YouTube)
Tested Amp Free in the CLI and VS Code extension on a solar panel website, found the free result decent and functional, likes the UI and finds the in-tool ads pretty good. Guesses complex work will still need the paid smart option.
"…this is working. For complex things, I guess you will going to need still the smart option to be enabled. But yeah, you need to add money. I'm curious to see if this uh plan it will work also the free one and the one…"
Practitioner account, self-described; direct use; counts toward Operational control, Cost predictability and limits
Positive
yomismoaqui, Hacker News commenter, self-described
Comment on: Claude Code for VSCodeHacker News comment
Spent about 15 euros, including 10 euros of free credit, trying Amp in one day and came away impressed.
Practitioner account, pseudonymous; direct use; not counted, takes no position
Positive
Ravindra Barthwal, founder building dxcode
Post to the Amp team on orbsX post
Found Amp good but costly in the past. After 4 days on the US$20 plan he gets a lot of value, with one orb working for over 4 days and coordinating with other orbs.
"In just $20 I am getting ton of value. One of my orb is working for over 4 days now and co-coordinating with other orbs…"
Practitioner account, pseudonymous; direct use; counts toward Reliability
Positive
Developer, formerly at Heptabase
AmpCode is coolPersonal blog post, blog.ianhsiao.me
Uses orbs and thread-to-thread messaging to move work between local and remote machines, and credits Amp for removing features as models improve.
Practitioner account, self-described; direct use; counts toward Adoption and upkeep
Positive
weiliddat, Hacker News commenter, self-described
Comment on: OpenAI will start testing ads in ChatGPT free and Go tiersHacker News comment
Uses Amp as their main coding agent outside work on the ad-supported free mode, which moved to a US$10 daily limit with Opus 4.5.
Practitioner account, pseudonymous; direct use; counts toward Cost predictability and limits
Positive
tbliu, Hacker News commenter, self-described
Comment on: Anthropic blocks third-party use of Claude Code subscriptionsHacker News comment
Uses Amp's free tier with Opus 4.5 heavily, calls it a very nice CLI, and finds the ads fairly unobtrusive.
Practitioner account, pseudonymous; direct use; counts toward Operational control, Cost predictability and limits
Positive
incoming1211, Hacker News commenter, self-described
Comment on: Amp, Inc. - Amp is spinning out of SourcegraphHacker News comment
Compares most terminal agents and says Amp costs more but gives the best results, crediting the Librarian and the Oracle.
Practitioner account, pseudonymous; direct use; not counted, repeats a stronger account
Positive
Independent developer
Amp: Notes from reading the Amp manualPersonal blog post, hamel.dev
Names threads their favorite Amp feature, finds model abstraction works despite early doubts, and pays per token because the time saved is worth it.
Practitioner account, self-described; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
Positive
Co-founder of Octoco
Getting Started with AmpPersonal blog post, G-J's Substack
A co-founder's starter guide for Octoco's senior engineers: keep threads short, plan with the Oracle, hand off phases, and expect cost to grow with long threads.
Practitioner account, self-described; direct use; counts toward Reliability, Operational control, Cost predictability and limits
Positive
Creator of Ghostty
Vibing a Non-Trivial Ghostty FeaturePersonal blog post, mitchellh.com
Built a Ghostty feature with Amp over 16 sessions totalling US$15.98 in token spend, consulting the Oracle for all planning. Guided the agent through cleanup sessions with minor manual changes and merged after a final manual review.
Practitioner account; direct use; counts toward Delivered outcomes, Operational control, Cost predictability and limits
Positive
Independent developer and contributor to the Air framework
Amp Code Deep Dive and TalkPersonal blog post, Elite AI Assisted Coding
Used Amp for a real open-source documentation task that was merged; it fixed a database bug from logs on the first try for US$4.10, though drafts needed trimming.
Practitioner account, self-described; structured test; counts toward Delivered outcomes, Operational control, Cost predictability and limits, Adoption and upkeep
GILD's fact checks, which don't count as sources
Checked Oct 12, 2026.Since September 13, 2026, Hobby is free with your own model keys or linked ChatGPT or Claude subscriptions, with no Amp token fees. Individual (Megawatt) is US$20 per month with orb time and credits. Paid credits use provider API prices with no markup outside Enterprise. Most cost complaints describe 2025 pay-as-you-go pricing. Pricing
Open them in the recordStatements this group's sources agree on, then the ones close to agreement.
Close to agreement: each needs 16 supporting sources
Practitioners use Amp as an agent that carries a task or a whole project to completion.
6 supporting sourcesQuestions the sources split on, with the count on each side.
Evidence remains mixed
Amp as main agent, 11 sources
Some keep Amp, or return to it, after trying Claude Code, Codex, and Cursor. They cite results and the feel of the tool.
Another agent as main, 9 sources
Others keep Claude Code, Codex, Devin, or Pi as their main agent. Cost or preference for another tool drives that choice.
Most sources say it is worth it
Worth it, 8 sources
Most sources that weigh price against results say time saved justifies Amp, from pay-per-token use to the US$20 monthly plan.
Not worth it, 3 sources
Others found 2025 pay-as-you-go spend too high against a flat Claude plan.
The three groups, then every practitioner by the context they work in, and how emphasis changes with scale.
Three groups drawn from the same sources. Choose a group to select whose experience counts. Within that group, each author counts once in the practitioner sentiment, with their latest view. Signals open on All practitioners. Leaders and teams have 6 accounts; GILD prints their sentiment from 20.
9 sources. These accounts focus on spend as usage spreads, shared threads, and what changes when Amp moves from one engineer to a team.
68 sources. These accounts focus on raw capability, cost per task, and how Amp performs in self-directed development.
77 sources. This gives the broadest market read, but combines work settings that can produce materially different conclusions.
77 sources from practitioners, by the context they work in. Filled rows count for this group.
Emphasizes Results and cost per task
Emphasizes Orbs and team spend
Emphasizes Review practice and workspace controls
Emphasizes How to introduce Amp and keep threads short
What changes with scale. Individual builders judge Amp on results, the Oracle, model routing, and task cost. The few team accounts focus on spend per engineer, shared threads, and moving agent runs off laptops onto orbs.
Where this group's sources were published.
Counted for All practitioners, from 77 sources. Change group
95 sources for this group, 18 of them documentation
Older evidence stays when it still applies. Accounts tied to changed plans, retired features, or earlier model lineups remain in the history with a Needs a recheck label.
See sources by originWho each source is, how it knows, whether it still applies and what it counts toward.
Newest first, on its own page. Every count above opens its sources there.
Read all 95 sourcesAbout this read
A running review of what engineering teams report about Amp, from individual builders to engineering leaders. Each finding links to its sources, and each conflict shows both sides.
Counts cover only the sources GILD reviewed.
Choose a group to select whose experience counts. Within that group, each author counts once in the practitioner sentiment, with their latest view. GILD files each source in a group from how the account describes its author. A role or employer marked self-described is what the source states about themselves. Documentation checks facts for every group and never counts toward sentiment.
A source's date and whether it still applies are separate: a recent post can describe an old version, and an old one can still hold.
Flagged for a recheck, of the sources that take a side: leaders and teams 3 of 6, individual builders 21 of 50, all practitioners 24 of 56.
Each row opens its sources for the group you are reading.
| Finding | Engineering leaders and teams | Individual builders | All practitioners | Sources |
|---|---|---|---|---|
| Current readCurrent sentiment | 5 positive, 0 mixed, 1 negative | 37 positive, 11 mixed, 2 negative | 42 positive, 11 mixed, 3 negative | Sources |
| PraiseOracle & Model Routing | 2 of 2 positive (under the minimum) | 12 of 12 positive | 14 of 14 positive | Sources |
| PraiseResult Quality | 1 of 1 positive (under the minimum) | 14 of 15 positive | 15 of 16 positive | Sources |
| PraiseDeveloper Experience | 1 of 2 positive (under the minimum) | 13 of 15 positive | 14 of 17 positive | Sources |
| PraiseThreads & Handoff | 2 of 2 positive (under the minimum) | 11 of 11 positive | 13 of 13 positive | Sources |
| PraiseOrbs & Cloud Agents | 2 of 2 positive (under the minimum) | 10 of 10 positive | 12 of 12 positive | Sources |
| PraiseValue for Money | 0 of 0 positive (under the minimum) | 3 of 9 positive | 3 of 9 positive | Sources |
| ConcernUsage / Cost | 0 of 2 raise it (under the minimum) | 11 of 24 raise it | 11 of 26 raise it | Sources |
| ConcernRemoved Features | 1 of 1 raise it (under the minimum) | 1 of 3 raise it (under the minimum) | 2 of 4 raise it (under the minimum) | Sources |
| ConcernRequires Strong Review | 0 of 0 raise it (under the minimum) | 1 of 3 raise it (under the minimum) | 1 of 3 raise it (under the minimum) | Sources |
| ConcernAutonomy & Permissions | 1 of 1 raise it (under the minimum) | 1 of 3 raise it (under the minimum) | 2 of 4 raise it (under the minimum) | Sources |
| ConcernNeeds Close Steering | 0 of 0 raise it (under the minimum) | 1 of 3 raise it (under the minimum) | 1 of 3 raise it (under the minimum) | Sources |
| ConsensusEvaluated as an engineering agent | 1 supporting (under the minimum) | 5 supporting (below threshold) | 6 supporting (below threshold) | Sources |
| ConsensusWritten instructions make results consistent | 1 supporting (under the minimum) | 3 supporting (under the minimum) | 4 supporting (under the minimum) | Sources |
| DisagreementAmong practitioners who compared agents, does Amp stay their main agent? | Amp as main agent 1, Another agent as main 1 (under the minimum) | Amp as main agent 10, Another agent as main 8 | Amp as main agent 11, Another agent as main 9 | Sources |
| DisagreementIs Amp worth what it costs? | Worth it 1, Not worth it 0 (under the minimum) | Worth it 7, Not worth it 3 | Worth it 8, Not worth it 3 | Sources |
| Current readEvidence mix | 27 sources | 86 sources | 95 sources | Sources |
GILD's checks of claims against official documentation, kept apart from the source counts.
