Skip to main content

Amp Frontier CorporationAI Coding Agent

Amp

Amp is a coding agent from Amp Frontier Corporation, spun out of Sourcegraph in December 2025. Engineers use the CLI, web, macOS, iOS, or Android apps. Threads run locally, on their own runner, or in an orb, a cloud machine Amp creates per thread. Amp chooses Anthropic, OpenAI, and other models per mode.

The 30-Second Take

Pick Amp if your engineers hand off whole tasks and run several agents at once.

It picks the models itself and sends plans to a second model, the Oracle, for a check. Threads run locally, on your own runner, or in a cloud machine Amp creates per thread.

Best fit
Teams that delegate whole tasks to agents on cloud machines and review what comes back.
Main constraint
No approval prompts by default, and per-user spend limits only on Enterprise.
First move
Install a workspace plugin that gates tool calls, set every project to push to a branch, and write a mode policy.

Why

Watch the defaults: agents run tools without approval unless you change that, and per-user spend limits come only on Enterprise. Amp Frontier Corporation spun out of Sourcegraph in December 2025, and it retires features quickly.

Practitioner sentiment, all practitioners+69 Strong positive

51 practitioners, 95 sources reviewed, Updated Nov 1, 2026How this is measured

Read the GILD Review
Save a product your organization uses.Keep up with Amp without adding it to your stack.Visit Website
Operating stack
Agentic Software Development

Comparisons

How does Amp compare with the alternatives?

  1. Amp vs Claude Code

    • AI Coding Agents

    Choose Claude Code for approvals on by default and VS Code and JetBrains extensions on Claude plans. Choose Amp for threads in cloud orbs on models from several vendors. Since October 2026, Amp also runs Claude Code on a Claude plan.

  2. Amp vs Cursor

    • AI Coding Agents

    Choose Cursor if engineers will work in one editor with cloud agents on the side. Choose Amp if they will steer threads from the CLI or web and run them in orbs. Then compare team spend controls, where Amp's per-user limits need Enterprise.

Operator Insights

What are engineering leaders learning from deploying Amp?

See all 7 operator accounts

Buying and deployment

What engineering leaders should verify before a broader rollout.

Sourced from Amp documentation

How you buy it

US$20

A month for Individual. Gigawatt adds the largest orb, Hobby is free, and Teams adds no charge.

Source 1, Amp, Pricing: PricingSource 5, Amp, Docs: Orb sizes and costs: Docs: Orb sizes and costs

Details

Hobby is free. Individual (Megawatt) costs US$20 per month with 45,000 minutes of time on Amp-hosted orbs; Gigawatt, the higher Individual tier, is the only one with the a1.3xlarge orb size. A Teams workspace adds no charge and accepts members on any tier. Enterprise has custom pricing with pooled credits and no paid seats.

Cost drivers

No markup

Inference bills at providers' API prices outside Enterprise. Orbs bill by the minute.

Source 2, Amp, Docs: Pricing: Docs: PricingSource 1, Amp, Pricing: PricingSource 5, Amp, Docs: Orb sizes and costs: Docs: Orb sizes and costs

Details

Inference bills to Amp credits at providers' API prices, with no markup outside Enterprise. It can also bill to your own keys or linked ChatGPT and Claude plans, with no Amp token fees. Orbs bill by the minute, from US$0.08 to US$2.13 per hour by size. Projects default to a1.medium at US$0.33 per hour, and personal projects on Megawatt default to a1.small at US$0.17.

Where it runs

CLI, apps, cloud

Threads run on your runners, Amp's orbs, or self-hosted orbs. The Amp service itself is hosted.

Source 3, Amp, Docs: Introduction: Docs: IntroductionSource 7, Amp, Docs: CLI: Docs: CLISource 8, ampcode.com, Amp news: Amp on Android (Oct 27, 2026): Amp news: Amp on Android (Oct 27, 2026)Source 9, ampcode.com, Amp news: Self-Hosted Orbs (Oct 22, 2026): Amp news: Self-Hosted Orbs (Oct 22, 2026)Source 5, Amp, Docs: Orb sizes and costs: Docs: Orb sizes and costsSource 11, Amp, Amp Security Reference: Amp Security ReferenceSource 10, Amp, Amp news: The Coding Agent Is Dead (Feb 19, 2026): Amp news: The Coding Agent Is Dead (Feb 19, 2026)

Details

Amp has a CLI for macOS, Linux, and Windows through WSL, plus web, macOS, iPhone and iPad, and Android apps; the Android app arrived on October 27, 2026. Threads run on your own runners, on orbs Amp hosts, or on self-hosted orbs, whose orb provider runs inside your infrastructure and reaches out to Amp. Self-hosted orbs are free on Hobby, Individual, and Teams, and Enterprise pays the standard orb price for them from October 29, 2026. The Amp service itself has no self-hosted deployment. Editor extensions ended March 5, 2026.

What to verify

No approval by default

Amp runs tools without asking, and the default Ship behavior pushes straight to the base branch.

Source 4, Amp, Docs: Tools: Docs: ToolsSource 6, ampcode.com, Docs: Shipping changes: Docs: Shipping changesSource 2, Amp, Docs: Pricing: Docs: Pricing

Details

Amp runs tools without asking. The default Ship behavior commits, rebases, and pushes straight to the base branch; set Ship Behavior to Push to Branch or a Custom Ship prompt to require pull requests. SCIM directory sync, MCP registry allowlists, and per-user cost controls require Enterprise.

Signals

What is the current evidence saying about Amp?

A current read of what engineering leaders, teams, and individual builders report publicly about Amp, from the retired editor extension to the CLI and orbs, with each finding linked to its sources.

Whose sources count

Practitioner sentiment, all practitioners

+69 Strong positive net

of 56 sources that take a side, each author counted once.

How this figure is made

20 more state no overall view and cast no vote.

Leaders and teams have 6 accounts; GILD prints their sentiment from 20.

Current sentiment

Positive, mixed and negative sources

The score counts them as 38 positive, 10 mixed and 3 negative votes, one per author.

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.

Use a task pilot to measure accepted work and review time.

Evidence field. All 95 sources, one mark each.

Arrange by

Each mark is one source: positive, mixed, negative, no opinion stated, or documentation (not counted).

By sentiment, All practitioners. Every source GILD read, grouped as the field draws it.
GroupSourcesPositiveMixedNegativeNo opinion stated and not counted in the scoreDocumentation, checked and not counted
Positive42420000
Mixed11011000
Negative300300
Not counted21000210
Documentation18000018
Every source GILD read, by the quarter it was published, as the field's time axis counts it. Documentation is dated by the day it was checked.
QuarterSources
Q2 20253
Q3 202516
Q4 202516
Q1 202616
Q2 20267
Q3 202618
Q4 20261
Documentation, checked18

By sentiment

18 documentation pages check facts and are not counted.

Hover a mark for a quick look, click it or press Enter to open its source.Tap a mark to open its source, then step to its neighbors with Previous and Next.

What practitioners praise

Top praise of 6. Each mark is a source on it; filled marks praise it.

  1. Result Quality15 of 16 sources on it praise it

Most common concerns

Each mark is a source on it; filled marks raise it.

  1. Usage / Cost11 of 26 sources on it raise it as a concern
All 95 sources behind these findings
  1. PricingAmp (Official), Product documentation, checked Oct 12, 2026
  2. Docs: PricingAmp (Official), Product documentation, checked Oct 12, 2026
  3. Docs: Sizes & CostsAmp (Official), Product documentation, checked Oct 12, 2026
  4. Modes & ModelsAmp (Official), Product documentation, checked Oct 12, 2026
  5. Docs: The DialAmp (Official), Product documentation, checked Oct 12, 2026
  6. Docs: IntroductionAmp (Official), Product documentation, checked Oct 12, 2026
  7. Docs: OrbsAmp (Official), Product documentation, checked Oct 12, 2026
  8. Docs: CLIAmp (Official), Product documentation, checked Oct 12, 2026
  9. Docs: RunnersAmp (Official), Product documentation, checked Oct 12, 2026
  10. Docs: ToolsAmp (Official), Product documentation, checked Oct 12, 2026
  11. Docs: AGENTS.mdAmp (Official), Product documentation, checked Oct 12, 2026
  12. Docs: ThreadsAmp (Official), Product documentation, checked Oct 12, 2026
  13. Docs: Workspace EntitlementsAmp (Official), Product documentation, checked Oct 12, 2026
  14. Docs: Minimal Data RetentionAmp (Official), Product documentation, checked Oct 12, 2026
  15. Security ReferenceAmp (Official), Product documentation, checked Oct 12, 2026
  16. News: Amp Frontier CorporationAmp (Official), Product documentation, checked Oct 12, 2026
  17. News: The Coding Agent Is DeadAmp (Official), Product documentation, checked Oct 12, 2026
  18. News: Free AgentAmp (Official), Product documentation, checked Oct 12, 2026
  19. Post on moving Omega development to the cloudTaylor Otwell, creator of Laravel (self-described: Laravel product work across 5 repositories with multiple agents), X post, published Oct 7, 2026
  20. Comment on: Plan mode is deadcavoirom, Hacker News commenter, self-described, Hacker News comment, published Sep 26, 2026
  21. Comment on: Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astrafishtoaster, Hacker News commenter (self-described: First engineer at an early-stage startup), Hacker News comment, published Sep 11, 2026
  22. Comment on: Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astramicrodrum, Hacker News commenter, self-described, Hacker News comment, published Sep 10, 2026
  23. 147: It's Orbin' TimeCompany founder, co-host of Mostly Technical (self-described: Their company's app work; the company's CTO tried orbs first), Podcast episode, Mostly Technical (podcast), published Sep 1, 2026
  24. Comment on: Orbsjosevalerio, Hacker News commenter, self-described, Hacker News comment, published Aug 30, 2026
  25. Comment on: Orbsaltschuler, Hacker News commenter, self-described, Hacker News comment, published Aug 30, 2026
  26. Comment on: Our decision on Cursor following its acquisition by SpaceXcavoirom, Hacker News commenter, self-described, Hacker News comment, published Aug 29, 2026
  27. Post on moving project work to AmpJoey Kudish, software developer with 20 years of experience (self-described: Independent developer running several products; also uses Codex), X post, published Aug 18, 2026
  28. Post on the team's use of Amp orbsJJ Englert, Enablement Lead at Tenex (self-described: AI consulting firm team moving work from local machines to the cloud), X post, published Aug 18, 2026
  29. Post on an agent recording a bug video in an orbAndrzej Krzywda, CEO of Arkency (Rails consultancy working on large legacy Rails applications), X post, published Aug 16, 2026
  30. Post to the Amp team on orbsRavindra Barthwal, founder building dxcode (self-described: Has tried Cursor, Devin, Codex, Claude, and other cloud agents), X post, published Aug 15, 2026
  31. Comment on: How Compaction Works in Pinavs, Hacker News commenter (self-described: Software engineer in Australia, New Zealand and Fiji), Hacker News comment, published Aug 14, 2026
  32. Comment on: Launch HN: Hoplite (YC S26) - Effortlessly deploy cloud coding agentsr5Khe, Hacker News commenter, self-described, Hacker News comment, published Aug 3, 2026
  33. The State of Agents in the CloudAndrei Maxim, applied AI engineer and former Ruby developer (self-described: Compared Amp orbs with Claude Code and Codex cloud environments on his own projects), X post (X article), published Aug 3, 2026
  34. AmpCode is coolDeveloper, formerly at Heptabase (self-described: Builds their own app; uses orbs from an iPad), Personal blog post, blog.ianhsiao.me, published Jul 31, 2026
  35. Amp Got Cloud Coding Sandboxes RightMike Hostetler, founder and Elixir agent framework builder (self-described: Maintains several open-source Elixir repositories, including ReqLLM), X post (X article), published Jul 20, 2026
  36. Comment on: Claude Code uses Bun written in Rust nowanamexis, Hacker News commenter, self-described, Hacker News comment, published Jul 19, 2026
  37. Comment on: Claude Sonnet 5usef-, Hacker News commenter, self-described, Hacker News comment, published Jul 1, 2026
  38. Comment on: GLM 5.2 beats Claude in our benchmarksnl, Hacker News commenter (self-described: Former ML startup founder and former startup CTO), Hacker News comment, published Jun 29, 2026
  39. Comment on: Sakana Fugunixosbestos, Hacker News commenter, self-described, Hacker News comment, published Jun 22, 2026
  40. Amp vs Claude Code: Is It Worth Switching in 2026?Software engineer (self-described: Solo build of a multi-agent setup with an MCP server and a web frontend), Personal blog post, alexdunlop.com, published Jun 15, 2026
  41. Comment on: Is Amp more or less expensive than Claude Code? Is it better?mutant, Hacker News commenter, self-described, Hacker News comment, published May 28, 2026
  42. Comment on: Elevated error rates on Opus 4.7SatvikBeri, Hacker News commenter, self-described, Hacker News comment, published May 15, 2026
  43. CI/CD Is Dead, Agents Need Continuous Compute and ComputersHugo Santos, Founder and CEO of Namespace (Namespace, a build and test compute infrastructure company whose team works from specs through agent loops), Conference talk page with transcript, AI Engineer Europe 2026, published May 13, 2026
  44. Comment on: DeepClaude - Claude Code agent loop with DeepSeek V4 Proahmadyan, Hacker News commenter (self-described: Maker of Agentastic), Hacker News comment, published May 4, 2026
  45. Post on migrating off AmpJo Kristian Bergum, CEO of Hornet (Hornet, a Trondheim startup building a retrieval engine for agents), X post, published Mar 25, 2026
  46. Comment on: How I write software with LLMsshell0x, Hacker News commenter, self-described, Hacker News comment, published Mar 16, 2026
  47. Comment on: Show HN: Praxis, my personal take on Compound Engineering with AIDFilipeS, Hacker News commenter, self-described, Hacker News comment, published Feb 24, 2026
  48. Comment on: Show HN: Tlsctl - A CLI for inspecting and troubleshooting TLS_smertens_, Hacker News commenter, self-described, Hacker News comment, published Feb 22, 2026
  49. Comment on: Claude Code's compaction discards data that's still on diskSatvikBeri, Hacker News commenter, self-described, Hacker News comment, published Feb 21, 2026
  50. Comment on: Amp Free Is Fulljauntywundrkind, Hacker News commenter, self-described, Hacker News comment, published Feb 11, 2026
  51. Comment on: Claude Code is being dumbed down?4b11b4, Hacker News commenter, self-described, Hacker News comment, published Feb 11, 2026
  52. Comment on: Review-for-agent: A local, PR-style UI for reviewing AI agent code changesasadjb, Hacker News commenter, self-described, Hacker News comment, published Feb 10, 2026
  53. I keep coming back to Amp CodeIndependent developer (self-described: Solo; also uses Claude Code (Max) and Codex), Personal blog post, Jibran's Perspective (asadjb.com), published Feb 8, 2026
  54. Should You Use Ampcode for Production Code? One Month InDeveloper (self-described: Healthcare software work; runs Amp on a dedicated branch), Personal blog post, IntelligentTools, published Feb 7, 2026
  55. Comment on: Claude Code: connect to a local model when your quota runs outzingar, Hacker News commenter, self-described, Hacker News comment, published Feb 4, 2026
  56. AMP vs Claude Code (CLI): Comparing of Agentic Coding ToolsWeb developer (self-described: Professional web development; over 95 percent of their code is AI-generated), Personal blog post, cesarcouto.com, published Jan 21, 2026
  57. Coding with Claude Opus 4.5 For Free! (AmpCode)Developer and YouTube creator (self-described: Solo), Video, YouTube, published Jan 19, 2026
  58. Comment on: OpenAI will start testing ads in ChatGPT free and Go tiersweiliddat, Hacker News commenter, self-described, Hacker News comment, published Jan 18, 2026
  59. Comment on: Unauthenticated remote code execution in OpenCodetheshrike79, Hacker News commenter, self-described, Hacker News comment, published Jan 12, 2026
  60. Comment on: Anthropic blocks third-party use of Claude Code subscriptionstbliu, Hacker News commenter, self-described, Hacker News comment, published Jan 9, 2026
  61. Comment on: Amp, Inc. - Amp is spinning out of Sourcegraphtruekonrads, Hacker News commenter, self-described, Hacker News comment, published Dec 9, 2025
  62. Comment on: Amp, Inc. - Amp is spinning out of Sourcegraphthe_mitsuhiko, Hacker News commenter, self-described, Hacker News comment, published Dec 9, 2025
  63. Comment on: Amp, Inc. - Amp is spinning out of Sourcegraphincoming1211, Hacker News commenter (self-described: Engineering team at work), Hacker News comment, published Dec 9, 2025
  64. Comment on: Amp, Inc. - Amp is spinning out of Sourcegraphsergiotapia, Hacker News commenter, self-described, Hacker News comment, published Dec 9, 2025
  65. Comment on: Amp, Inc. - Amp is spinning out of Sourcegraphlvl155, Hacker News commenter, self-described, Hacker News comment, published Dec 8, 2025
  66. Comment on: Amp, Inc. - Amp is spinning out of Sourcegraphembedding-shape, Hacker News commenter, self-described, Hacker News comment, published Dec 8, 2025
  67. Comment on: Amp, Inc. - Amp is spinning out of Sourcegraphincoming1211, Hacker News commenter, self-described, Hacker News comment, published Dec 8, 2025
  68. Comment on: Agent design is still hardpjm331, Hacker News commenter, self-described, Hacker News comment, published Nov 22, 2025
  69. Comment on: You should still be writing code from your editorMorehouseJ09, Founder and CEO of Nuon, Hacker News commenter (self-described: Founder and CEO of Nuon), Hacker News comment, published Nov 16, 2025
  70. Amp Coding Agent: Features That Make It Stand OutIndependent developer (self-described: Engineering lead for CI/CD and DevOps), Personal blog post, peterwarnock.com, published Nov 10, 2025
  71. Amp: Notes from reading the Amp manualIndependent developer (self-described: Solo use, including SEO work on their own site), Personal blog post, hamel.dev, published Nov 2, 2025
  72. Getting Started with AmpCo-founder of Octoco (self-described: Octoco, senior engineering staff being introduced to Amp), Personal blog post, G-J's Substack, published Oct 28, 2025
  73. Comment on: Claude Code on the webhnidiots3, Hacker News commenter, self-described, Hacker News comment, published Oct 20, 2025
  74. I Tried the Ad-Supported AI Coding Agent (Amp Free)WEBdoze, YouTube channel (self-described: Solo, small web projects), Video, WEBdoze (YouTube), published Oct 17, 2025
  75. Vibing a Non-Trivial Ghostty FeatureCreator of Ghostty (self-described: Ghostty, open-source terminal emulator; macOS update notification feature), Personal blog post, mitchellh.com, published Oct 11, 2025
  76. Amp - Vibe Code With Me - Ep 016Developer (self-described: Solo, building an expense app on stream), Video, YouTube, published Oct 6, 2025
  77. Comment on: Ask HN: What are you working on? (September 2025)jeanlucas, Hacker News commenter, self-described, Hacker News comment, published Sep 30, 2025
  78. Changelog #658: Adam Stacoviak on running agents side by sideAdam Stacoviak, co-founder and podcast host, Changelog (self-described: Changelog, his own long-running project), Podcast episode with publisher transcript, published Sep 17, 2025
  79. Can AMP Code Beat Claude Code? | Feature Build Day 3Developer and streamer (self-described: Solo, building a photo upload feature for their own app), Video (livestream), YouTube, published Sep 7, 2025
  80. Post on shared Amp threads in pull requestsMitchell Hashimoto, creator of Ghostty and co-founder of HashiCorp (self-described: Maintainer of the Ghostty terminal, reviewing outside pull requests), X post, published Sep 3, 2025
  81. Comment on: Ask HN: What LLM are you all using for coding assistance right now?aitooltrek-com, Hacker News commenter (self-described: Indie maker building an AI tools site), Hacker News comment, published Aug 13, 2025
  82. Comment on: Getting good results from Claude Codederencius, Hacker News commenter, self-described, Hacker News comment, published Aug 9, 2025
  83. Comment on: Jules, our asynchronous coding agentpjm331, Hacker News commenter, self-described, Hacker News comment, published Aug 6, 2025
  84. Comment on: Claude Opus 4.1mark_undoio, CTO of Undo, Hacker News commenter (self-described: CTO of Undo, a record and replay debugging company), Hacker News comment, published Aug 5, 2025
  85. Amp Code: Arbitrary Command Execution via Prompt Injection FixedJohann Rehberger, security researcher (Embrace The Red) (self-described: Independent security testing of coding agents during his Month of AI Bugs series), Independent security evaluation, Embrace The Red, published Aug 5, 2025
  86. Comment on: 6 weeks of Claude Codesergiotapia, Hacker News commenter, self-described, Hacker News comment, published Aug 3, 2025
  87. Amp Code Deep Dive and TalkIndependent developer and contributor to the Air framework (self-described: Open-source documentation task for the Air web framework), Personal blog post, Elite AI Assisted Coding, published Jul 30, 2025
  88. AI coding agent one-shot challenge with Amp, Claude Code and GeminiIndependent developer maintaining open-source side projects (self-described: Solo, side projects), Personal blog post, Land of Unknwon (unknwon.io), published Jul 22, 2025
  89. Comment on: Ask HN: Why do Cursor, Windsurf and Claude Code dominate the conversation?Gys, Hacker News commenter, self-described, Hacker News comment, published Jul 21, 2025
  90. Comment on: Cognition (Devin AI) to Acquire Windsurfd_sc, Hacker News commenter, self-described, Hacker News comment, published Jul 16, 2025
  91. Comment on: AI slows down open source developers. Peter Naur can teach us whyyomismoaqui, Hacker News commenter, self-described, Hacker News comment, published Jul 14, 2025
  92. Comment on: Clarifying our pricingmrklol, Hacker News commenter, self-described, Hacker News comment, published Jul 5, 2025
  93. Comment on: Claude Code for VSCodeyomismoaqui, Hacker News commenter, self-described, Hacker News comment, published Jun 23, 2025
  94. Transcript: 'How Two Engineers Ship Like a Team of 15 With AI Agents'Kieran Klaassen, general manager of Cora at Every (Two-engineer team building Cora, an AI email product at Every, running several coding agents side by side), Podcast episode transcript, AI & I (Every), published Jun 11, 2025
  95. Comment on: Cursor 1.0vladstudio, Hacker News commenter (self-described: Self-employed wallpaper designer), Hacker News comment, published Jun 5, 2025

Jun 2025 to Oct 2026.

77 sources from practitioners, by month and sentiment.

The full readExplore all Amp Signals6 praise themes, 5 concerns, 2 points of agreement and 2 disagreements across all three groups, with all 95 sources open to inspect.

Where It Fits

Where does Amp belong in the stack?

Open Where It Fits: the Stack Field and the full map

Amp is Core across Build, supports Plan, Validate, and Ship, and touches Operate.

Business area
Product & Engineering
Function
Software Development
Operating stack
Agentic Software Development
Primary role
Coding Agent & Cloud Dev Environment
Amp beside the rest of its stackThe Stack Field, every product on every layer
FootprintAmp across the Agentic Software Development stack. Bar height is its role on each layer.

  1. Plan

    1. Work planning: Supports

      Work planning, supports. The Oracle plans hard changes, and the Task subagent handles delegated work. Puck coordinates threads across projects. The team's issue tracker remains the planning system of record.

  2. Build

    1. Codebase context: Supports

      Codebase context, supports. Search retrieves code within a thread. Librarian researches external and connected GitHub repositories. Amp keeps no persistent codebase index, so a dedicated code search product owns that role.

    2. Coding: Core

      Coding, core. Amp edits a repository in low, medium, high, or ultra mode. Each mode pairs a main model with the Oracle and subagents Amp selects.

    3. Agent runtime: Core

      Agent runtime, core. Engineers steer Amp from the CLI, web, macOS, iOS, or Android. Threads run locally, on a runner, or in an orb, with shared threads and multiplayer review.

    4. Tool execution: Supports

      Tool execution, supports. Amp runs shell commands and MCP servers without approval prompts by default, as it does plugin tools. Orbs mint OIDC tokens for cloud access. Integration platforms and policy engines remain separate.

  3. Validate

    1. Debugging: Supports

      Debugging, supports. The Oracle reviews and debugs with a second frontier model. Orbs can read production logs through OIDC grants. Amp does not replace a debugging or incident tool.

    2. Testing: Supports

      Testing, supports. Orbs run test suites and serve the app through portals. The agent can show screenshots and running builds, while dedicated test infrastructure remains.

    3. Code review: Supports

      Code review, supports. Amp's Changes screen orders diffs, and the Oracle can review them. The engineering team still owns pull request review and merge decisions.

    4. Application security: Touches

      Application security, touches. Amp redacts secrets from threads and orb output and supports passkey-protected sessions. It does not scan code for vulnerabilities.

  4. Ship

    1. Source control: Supports

      Source control, supports. Amp pushes branches and opens pull requests through its GitHub app. It also restacks changes into commits and can host a project's repository. Most teams keep GitHub or GitLab as the system of record.

    2. CI/CD: Touches

      CI/CD, touches. Ship Changes tells the agent to run tests before rebasing and pushing. Orbs can watch a rollout. The team's CI/CD platform runs the delivery pipeline.

  5. Operate

    1. Observability: Touches

      Observability, touches. Scheduled and event-driven orbs can inspect production metrics and logs the team grants them. Amp provides no monitoring or alerting.

Market mapThe layers Amp covers, and who else owns each one.

  1. Coding & implementation

    CoreBuild

  2. Agent interface & developer runtime

    CoreBuild

  3. Work planning & task decomposition

    SupportsPlan

  4. Codebase context & intelligence

    SupportsBuild

  5. Tool execution & integrations

    SupportsBuild

7 more layers on the full map. Core, Supports and Touches describe coverage. They say nothing about quality.

More on Amp

  1. GILD ReviewGILD's evaluation, with the buying facts
  2. Research2 pieces
  3. Guides3 guides
  4. Operator Insights7 operator accounts
  5. Where It FitsOn the Agentic Software Development stack
  6. Related Products8 products