The second bill: what Microsoft 365 Copilot, GitHub Copilot, Amazon Q Developer and Salesforce Agentforce charge after the seat price
· 15 min read

The second bill: what Microsoft 365 Copilot, GitHub Copilot, Amazon Q Developer and Salesforce Agentforce charge after the seat price

Microsoft Copilot, GitHub Copilot, Amazon Q Developer and Salesforce Agentforce meter usage beyond the seat price, and this guide shows where to cap spend.

Every AI tool in this guide has a seat price, and every one of them now has a second invoice. Microsoft 365 Copilot, GitHub Copilot, Amazon Q Developer and Salesforce Agentforce all meter usage on top of the per-seat fee. Here is how each metering model works, what a 100-person deployment costs under each one, and the exact admin setting that caps spend.

What is “the second bill” on AI tools?

The first bill is the subscription you budget for: a fixed price per user per month. The second bill is usage-based charging that arrives on top of it once your team actually runs AI work. Four terms cover almost everything:

  • A credit is a vendor’s unit of account for AI work. You are granted a number of credits, or you buy them, and features deduct credits as you use them.
  • An action is one discrete task an AI agent performs. Salesforce defines it as a specific function such as updating a record, summarizing a complex case, answering a product inquiry, or executing a custom prompt or flow (Salesforce).
  • A pooled allowance is a shared pot of included units across the whole deployment, rather than separate per-person buckets.
  • Overage is the charge that applies once usage exceeds the included allowance.

Microsoft frames the split directly: “Everyday AI is included in your license. Advanced AI—work you hand off—is billed as you use it” (Microsoft). Its pricing page carries a blunt footnote: “Some AI experiences require usage-based billing and may result in charges based on usage, beyond your license fees” (Microsoft). All four vendors below have some version of that second line item.

The four metering models, in plain English

Microsoft 365 Copilot: credits with no published price. Microsoft charges for actual usage “measured in Copilot Credits” and offers fixed licensing, a Copilot Credit Pre-purchase Plan (P3), pay-as-you-go billing, prepaid capacity packs, or supported combinations of these (Microsoft Learn). Usage-based billing covers Cowork, advanced work in SharePoint, advanced work in OneDrive, the Work IQ API for custom apps and agents, Copilot Managed Runtime, Teams Phone Agent, and Code (Microsoft Learn). Critically, none of the Microsoft pages reviewed here publishes a per-credit dollar price (Microsoft Learn), so the metered layer cannot be pre-computed from the public pricing pages. For seats: Business Premium with Copilot is $32.00 user/month billed yearly or $38.40 on monthly billing; Business Standard with Copilot is $23.50 yearly or $28.20 monthly; the standalone Copilot add-on starts from $18.00, originally starting from $21.00 (Microsoft). The “Plans with Teams” versions of these bundles note that a Microsoft Teams license is required (Microsoft).

GitHub Copilot: credits with a fixed dollar value. Copilot Business costs $19 per granted seat per month and includes 1,900 AI credits per user per month; Copilot Enterprise costs $39 with 3,900 credits (GitHub). One AI credit equals $0.01, with credits derived from per-model token pricing (GitHub). Credit-consuming features are Copilot Chat, Copilot CLI, the Copilot cloud agent, Copilot Spaces, Spark, and third-party coding agents (GitHub). Inline code completions and next edit suggestions are not billed in AI credits and remain unlimited (GitHub). Licenses contribute credits to a shared pool (GitHub); Copilot access is a $0 seat for licensed GitHub Enterprise Cloud users, Copilot Standalone is $10 with no included allowance, and an enterprise must choose one billing model (GitHub).

Amazon Q Developer: seats plus metered transformation lines. The free tier includes 50 agentic requests per month; the Pro tier at $19 per user per month lists agentic requests as “Included (with limits)” — AWS publishes no number for that allowance, and we won’t guess one (AWS). The published meter is the Java transformation capability: free includes 1,000 lines of code per month, Pro includes 4,000 lines per user per month pooled at account level, and extra lines cost $0.003 per line submitted (AWS). The metering is fair: only submissions where you receive suggested upgrade code back are counted, jobs you stop or that fail are excluded, and non-code lines such as comments and empty lines don’t count (AWS). Pro adds an admin dashboard with user and policy management, automatic opt-out of data collection, and IP indemnity; the free tier has no dashboard, opt-out only, and no indemnity (AWS).

Salesforce Agentforce: credits bought in bulk, spent per action. Flex Credits cost $500 per 100,000 credits (Salesforce). Each action draws from that pool: standard Agentforce actions cost 20 Flex Credits, Agentforce Voice actions cost 30, and credits are fungible across Actions, Prompts, Translations and Voice Actions (Salesforce). The alternative is conversations at $2 per conversation (Salesforce). Licenses: the Agentforce User License is $5 user/month and requires Flex Credits; the Coworker User License is $50 with 30 MB of indexing per user per month; an Industries add-on at $150 user/month includes unmetered Agentforce usage for employees; Max edition starts from $550 user/month with the Agentforce add-on included and 2.75M Flex Credits per org per year; Salesforce Foundations is $0 (Salesforce).

The four meters side by side

ProductSeat priceMeterUnit price or allowanceNot metered / not countedSpend control
Microsoft 365 Copilot$18 add-on (was $21); $23.50–$32 bundled yearly (Microsoft)Copilot Credits (Microsoft Learn)No per-credit price on the pages reviewed (Microsoft Learn)“Everyday AI” included in the license (Microsoft)Spending policies and limits in the M365 admin center; “auto-apply new services” is on by default (Microsoft Learn)
GitHub Copilot$19 Business; $39 Enterprise (GitHub)AI credits; 1 credit = $0.01 (GitHub)1,900 / 3,900 credits per user per month, pooled (GitHub)Inline completions and next edit suggestions (GitHub)Disable the AI credits paid usage policy; enforced budgets with stop-usage (GitHub, GitHub)
Amazon Q DeveloperFree; $19 Pro (AWS)Lines of code (Java transformation) (AWS)4,000 lines per user per month pooled; extras at $0.003 per line (AWS)Stopped or failed jobs; comments and empty lines (AWS)No published hard cap; Pro admin dashboard manages users and policies (AWS)
Salesforce Agentforce$5 user license; $50 Coworker; $150 Industries; from $550 Max (Salesforce)Flex Credits at $500 per 100,000 (Salesforce)20 credits per action, 30 per Voice action; Max includes 2.75M credits per org per year (Salesforce)Unmetered employee usage on the Industries add-on (Salesforce)Digital Wallet monitoring with threshold alerts (Salesforce)

What 100 people actually cost

We model a 100-person deployment. Usage levels are stated per vendor, vendor-published figures are linked, and anything we calculated is labelled our arithmetic.

GitHub Copilot. Usage level: the team consumes the pool, then exceeds it. Seats: our arithmetic, $19 × 100 = $1,900 per month on Business (GitHub). GitHub documents that an enterprise with 100 Copilot Business users gets a shared pool of 190,000 AI credits rather than 100 individual buckets (GitHub). If the org consumes 300,000 credits in a month, our arithmetic: 110,000 excess credits × $0.01 = $1,100 in overage, so $3,000 all-in (GitHub). The included credits are worth exactly the seat price in metered terms — our arithmetic: 1,900 × $0.01 = $19.00 and 3,900 × $0.01 = $39.00 (GitHub, GitHub) — so the allowance is a bundled prepayment, not a bonus. Unused credits are forfeited, and the pool resets to the full monthly amount at 00:00:00 UTC on the first day of each calendar month (GitHub). Adding licenses mid-cycle grows the pool immediately; removing licenses doesn’t shrink it until the next billing cycle (GitHub).

Microsoft 365 Copilot. Usage level: any — the metered layer has no public unit price, so the second bill cannot be stated in dollars (Microsoft Learn). Seats only, our arithmetic on vendor prices: Business Standard with Copilot at $23.50 yearly billing is $23.50 × 100 × 12 = $28,200 per year; at $28.20 monthly billing it is $33,840 — a $5,640 difference, 20% more for monthly terms (Microsoft). Business Premium works out to $38,400 yearly ($32 × 100 × 12) or $46,080 on monthly billing ($38.40 × 100 × 12) (Microsoft). If a rollout goes wrong, you have a 7-day cancellation window after the paid subscription begins for a prorated refund, paying only for what you use (Microsoft).

Amazon Q Developer. Usage level: a transformation-heavy month with 500,000 lines of code submitted. Seats: our arithmetic, $19 × 100 = $1,900 per month (AWS) — consistent with the vendor’s own example of $190 for 10 users (AWS). At 100 active subscriptions, the pooled Pro allowance is 400,000 lines per month (AWS). Our arithmetic: 100,000 lines over × $0.003 = $300 in overage, so $2,200 for the month (AWS). Only successful submissions count toward that bill (AWS). Two cautions: the first month is pro-rated, but cancelling a user before month end still charges the full month’s fee; and the fee starts when someone actually requests work — an agentic coding plan, a transformation plan, or code completions — so a seat can sit unused at no charge and then activate on the first real request. Signing in to Amazon Q chat or downloading the toolkit does not start the fee (AWS). The vendor’s own daily-cost example works out to $6.33 per user at 0.333 of a month (AWS).

Salesforce Agentforce. Usage level: the vendor’s worked example — 60 Flex Credits × 3 cases a day × 20 days a month × 100 users, which Salesforce prices at $1,800 per month (Salesforce). Our arithmetic: that is 360,000 credits at $500 per 100,000, i.e. $0.005 per credit, $0.10 per standard action (20 credits) and $0.15 per Voice action (30 credits) (Salesforce). Add 100 user licenses at $5 each — our arithmetic: $500 — and the scenario totals $2,300 before other consumption services; Salesforce’s own disclaimer notes its examples exclude costs like Data 360 credits and that the page is subject to change (Salesforce). The conversation alternative: our arithmetic, $2 per conversation matches 20 metered actions, so conversations are cheaper when one averages more than ~20 actions, and credits are cheaper below that (Salesforce). A smaller vendor example — 20 credits × 5 questions a month × 20 new employees = $10 per month — shows how cheaply small deployments can run (Salesforce).

The admin switch that stops the second invoice

GitHub Copilot — the only documented default-on meter, with a named switch. On Copilot Business and Enterprise, additional usage is enabled by default; to prevent any spending beyond included AI credits, an administrator must explicitly disable the AI credits paid usage policy (GitHub). Budgets add a second layer: user-level budgets always enforce a hard stop, and a $0 user-level budget immediately blocks credit-consuming features (GitHub). But “stop usage when budget limit is reached” is off by default and applies only to enterprise spending limits, cost center budgets and organization budgets — without it, charges continue to accrue past the limit (GitHub). An enforced enterprise budget also does not cap license fees: our arithmetic restates GitHub’s example — 400 Business licenses at $7,600 plus an enforced $5,000 budget permits $12,600 combined, not $5,000 (GitHub). GitHub applies a “lowest remaining headroom wins” rule: whichever budget has the least capacity left blocks the user first, and a cost center with 10 Business and 5 Enterprise licenses caps at 38,500 credits (GitHub). If you hit an additional-usage cap, you must pay off what you have already consumed to continue (GitHub). One quirk: if a user holds licenses from multiple organizations, GitHub picks one organization at random each billing cycle to bill the seat (GitHub).

Microsoft 365 Copilot — policies and limits, plus one default to check. The cost management dashboard in the Microsoft 365 admin center lets admins set spending policies and limits, scope access by identity, configure alerts and user threshold notifications, manage billing methods, and review consumption by policy, user, group, agent, service and funding source (Microsoft Learn). Two honest caveats. First, “the ‘Auto-apply new services’ setting is enabled by default for spending policies,” so new metered services join your policy automatically until you change it (Microsoft Learn). Second, the Copilot Cost Management API reports usage in AI credits only — no pricing, invoices, currency costs or purchasing forecasts (Microsoft Learn). The documentation describes policies, limits and alerts rather than a single named kill switch, so set spending policies explicitly and turn off auto-apply on day one.

Amazon Q Developer — manage seats; no published dollar cap. The pricing page documents no hard spend-cap switch. What exists is the Pro tier’s admin dashboard with user and policy management, absent on the free tier (AWS). The practical levers are seat management and knowing the billing rules: the full-month charge on mid-month cancellation (AWS), and usage limits that “might be updated depending on regional factors, payment history, fraudulent usage, or approval of a quota increase request” (AWS).

Salesforce Agentforce — alerts plus structural caps. The Digital Wallet lets you monitor Flex Credit consumption in near real time, with proactive threshold alerts and usage trends (Salesforce). The pricing page documents monitoring and alerts, not a hard stop. Structural options exist: the Industries add-on at $150 user/month includes unmetered Agentforce usage for employees, and Max edition from $550 user/month includes 2.75M Flex Credits per org per year (Salesforce).

Which vendor is most predictable?

  1. GitHub Copilot. It publishes a fixed credit-to-dollar conversion (1 credit = $0.01 (GitHub)), published allowances (GitHub), a named default-on switch to disable (GitHub), hard-stop user budgets (GitHub), and unlimited completions (GitHub). The surprise risk is self-inflicted: overage is on by default and credits are forfeited monthly.
  2. Salesforce Agentforce. Every unit has a published price and the vendor publishes worked examples (Salesforce). The complexity is choosing credits versus conversations, and the documented control is alerts rather than a hard stop (Salesforce).
  3. Amazon Q Developer. The transformation meter is transparent — a pooled allowance and $0.003 per extra line (AWS) — but the Pro agentic-request allowance is unpublished (AWS), so the main chat meter can’t be forecast, and no hard cap is documented.
  4. Microsoft 365 Copilot. No per-credit price appears on the Microsoft pages reviewed (Microsoft Learn), so the second bill can’t be computed in advance, and the auto-apply default widens the metered surface (Microsoft Learn). The 7-day prorated cancellation window limits a bad first month (Microsoft).

FAQ: the second bill, answered

Does the seat price include unlimited AI usage?

Not for the heavy features. Microsoft includes “everyday AI” in the license while advanced hand-off work is billed as used (Microsoft). GitHub keeps inline completions and next edit suggestions unmetered, but Chat, CLI, the cloud agent, Spaces, Spark and third-party coding agents draw credits (GitHub, GitHub). Amazon Q’s Pro tier includes agentic requests “with limits” (AWS). Salesforce’s $5 and $50 licenses explicitly require Flex Credits (Salesforce).

What happens when we exceed GitHub Copilot’s credit pool?

Usage beyond the pool is charged at $0.01 per credit, and that additional usage is enabled by default on Business and Enterprise (GitHub, GitHub). Additional usage may be capped: if you hit the cap, you must pay off what you have already consumed to continue (GitHub).

Do unused AI credits roll over?

On GitHub, no. Included AI credits do not carry over between months; unused credits are forfeited, and the pool resets to the full monthly amount at 00:00:00 UTC on the first day of each calendar month (GitHub).

Can Amazon Q charge us after we cancel a user?

Yes. Cancel a user’s subscription before the end of the month and you are charged the full month’s fee; the first month is pro-rated (AWS). Activation is triggered by use rather than by sign-in: the fee starts when a user requests an agentic coding plan, a transformation plan, or code completions, while signing in to Amazon Q chat or downloading the toolkit does not start it (AWS).

Are Salesforce conversations or Flex Credits cheaper?

Our arithmetic: $2 per conversation matches 20 metered actions, so conversations win when a conversation averages more than ~20 actions, and credits win below that (Salesforce). Note that the vendor’s examples exclude other costs such as Data 360 credits (Salesforce).

Can we set a hard dollar cap on Microsoft 365 Copilot?

The documented controls are spending policies and limits, identity scoping, alerts and user threshold notifications in the admin center — with “auto-apply new services” enabled by default (Microsoft Learn). Since the reviewed Microsoft pages list no per-credit price, set limits in credit terms and watch the dashboard (Microsoft Learn).

The bottom line

GitHub Copilot is the most predictable of the four: it is the only vendor that publishes a fixed credit-to-dollar conversion, a named default-on spend switch you can disable, and budgets with a true hard stop (GitHub, GitHub). Salesforce is a close second on unit-price transparency (Salesforce); Microsoft is the hardest to forecast because its credit price isn’t published (Microsoft Learn). The one action to take before signing: find the exact spend-control setting for the product you are buying and set it during deployment, not after the first surprise invoice. For side-by-side verdicts on more AI tools, see our comparisons hub at toolbrain.net/comparisons.

How this guide was built

This guide is built from official pricing and documentation pages; we did not run these tools hands-on. Every figure links inline to its source: Microsoft’s pricing page and usage-based billing documentation, GitHub’s plans, billing and budgets pages, Amazon Q Developer pricing, and Salesforce Agentforce pricing. Where the pages reviewed publish no number — Microsoft’s per-credit price, Amazon Q’s Pro agentic-request allowance — we say so rather than estimate. Calculations marked “our arithmetic” multiply published prices by stated usage levels. Salesforce’s page notes that its examples exclude other consumption costs and are provided for information purposes only (Salesforce).

  • ToolBrain — tool reviews, LLM comparisons, and AI workflow guides
  • NiteAgent — AI agent development, frameworks, and production patterns

Cross-links automatically generated from None.

Back to all posts