Skip to main content

Buyer guide

Amp Team Rollout and Spend Checklist

What to settle before engineers move from personal Amp accounts to a shared workspace, and how to grade the pilot when it ends.

Reading time
2 min read
Sources
13 cited
In this piece
4 products

Judge an Amp team pilot by cost per accepted change against the agent you use today. Before it starts, choose a tier for each member, decide whose tokens pay for inference, budget orb minutes, and set spend controls.

When you need this

  • Engineers already use Amp on personal accounts and want a shared workspace.
  • You are choosing between Amp and Claude Code, Cursor, or Codex for the same team.
  • Finance wants a monthly ceiling before agents run on cloud machines.

Evaluation criteria

  1. Tier per member

    Which tier does each pilot member need?

    Look forTeams workspaces add no charge, and members can use different tiers. Hobby is free with pay-as-you-go orbs or your own runners. Individual (Megawatt) costs US$20 per month with 45,000 orb minutes and included credits. Gigawatt is the only self-serve tier with the a1.3xlarge orb. Enterprise is custom, with pooled credits and no paid seats.

  2. Whose tokens pay

    Will inference bill to Amp credits, your own keys, or linked subscriptions?

    Look forChoose which account pays for inference. Amp credits use provider API prices with no markup for individuals and non-enterprise workspaces. Since September 13, 2026, bringing your own keys or linking a ChatGPT or Claude subscription carries no Amp token fees. ChatGPT plans cover OpenAI roles in each mode. Under Amp's default modes, roles that run other providers' models, such as the medium main agent on Opus 5.5 and the low main agent on GLM-5.3-Flash, still bill credits. Linking ChatGPT applies a ChatGPT Only preset that pins the low, medium, and high main agents to GPT-5.6 Terra, GPT-6.1 Sol, and GPT-6 Astra, unless the modes were already tuned. A Claude plan covers only Claude Code mode.

  3. Orb minutes

    What will cloud machines cost at your usage?

    Look forBudget orbs 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. The default project size, a1.medium, costs US$0.33 per hour, and personal projects on Megawatt default to a1.small at US$0.17. Your own runners cost nothing, and a shared runner serves the workspace.

  4. Spend controls

    What stops one engineer's usage from surprising finance?

    Look forEnterprise adds per-user entitlements such as US$50 a week or US$100 a day. They count bring-your-own-key spend at estimated list price but not linked subscriptions. Below Enterprise, credits pool in the workspace. Set a mode policy and review usage weekly. Amp shows thread cost and the amp usage report shows the balance.

  5. Accepted-change worksheet

    What will you compare when the pilot ends?

    Look forTrack merged changes, review time, post-merge defects, credits, orb minutes, and spend per accepted change per engineer and week. Compare with your current agent on the same tasks. Public accounts range from a Ghostty feature at US$15.98 over 16 sessions to more than US$1,000 a month per engineer. Measure your own workload.

Questions to ask vendors

  1. Which members need Gigawatt, and which can stay on Hobby with their own subscriptions?
  2. Do we want Enterprise spend limits, or can we manage pooled credits without them?
  3. Which orb size should each project default to, and who changes it?
  4. Does our contract fix prices through renewal, given four pricing models since 2025?

Red flags

  • A pilot judged on threads started when it should be judged on changes accepted.
  • Every project on a large orb size by default.
  • Linked personal subscriptions used for company code without checking the provider's training settings.
  • No owner for the workspace's model routing and mode policy.

Worth evaluating

Listed in no particular order.

Sources

Official documentation
9
Public source
4
  1. Amp pricingOfficial documentationampcode.com
  2. Amp docs: PricingOfficial documentationampcode.com
  3. Amp docs: Orb sizes & costsOfficial documentationampcode.com
  4. Amp docs: The DialOfficial documentationampcode.com
  5. Amp docs: Model routingOfficial documentationampcode.com
  6. Amp docs: WorkspacesOfficial documentationampcode.com
  7. Amp docs: Workspace entitlementsOfficial documentationampcode.com
  8. Amp docs: RunnersOfficial documentationampcode.com
  9. Amp news: Free Agent (Sep 13, 2026)Official documentationampcode.com
  10. Mitchell Hashimoto: Vibing a Non-Trivial Ghostty Feature (Oct 11, 2025)Public sourcemitchellh.com
  11. Hacker News: a team spending more than US$1,000 a month each on Amp (Dec 9, 2025)Public sourcenews.ycombinator.com
  12. Hacker News: orbs and the US$20 plan (Aug 30, 2026)Public sourcenews.ycombinator.com
  13. Jibran: I keep coming back to Amp Code (Feb 8, 2026)Public sourceasadjb.com

Keep reading

  1. Buyer guideCodeRabbit in Your Coding Agents' LoopHow to wire the CodeRabbit CLI into Claude Code, Codex, Cursor, or Gemini, for teams whose agents open pull requests faster than reviewers read them.
  2. Buyer guideDelegating Linear Issues to Coding AgentsRoute each issue to an agent, write a usable brief, set access and AI credit limits, and require a human review before any merge.
  3. Buyer guideHow to Evaluate an AI Coding Agent Before Team RolloutHow to run a pilot that measures a coding agent's output quality, review burden, workflow fit, cost, and governance.
Read the research