From 2033d013590c99f73ffee5965fed3a147072d470 Mon Sep 17 00:00:00 2001 From: "well-architected-sync-bot[bot]" <235114805+well-architected-sync-bot[bot]@users.noreply.github.com> Date: Thu, 14 May 2026 14:26:15 +0000 Subject: [PATCH] Sync from github/github-well-architected-internal (main) Source Repository: github/github-well-architected-internal Source Branch: main Source SHA: 11d60a545b8698a9532053365429cd3b789a72c7 --- .../copilot_pru_enterprise_admin_playbook.md | 653 ------------------ .../recommendations/governing-agents.md | 4 +- .../recommendations/managing-ai-credits.md | 615 +++++++++++++++++ .../library/overview/release-notes/2025-q3.md | 2 +- docs/taxonomies.md | 2 +- 5 files changed, 619 insertions(+), 657 deletions(-) delete mode 100644 content/library/governance/recommendations/copilot-policies-best-practices/copilot_pru_enterprise_admin_playbook.md create mode 100644 content/library/governance/recommendations/managing-ai-credits.md diff --git a/content/library/governance/recommendations/copilot-policies-best-practices/copilot_pru_enterprise_admin_playbook.md b/content/library/governance/recommendations/copilot-policies-best-practices/copilot_pru_enterprise_admin_playbook.md deleted file mode 100644 index b4ad920..0000000 --- a/content/library/governance/recommendations/copilot-policies-best-practices/copilot_pru_enterprise_admin_playbook.md +++ /dev/null @@ -1,653 +0,0 @@ ---- -# SPDX-FileCopyrightText: GitHub and The Project Authors -# SPDX-License-Identifier: MIT -draft: false -title: 'Managing GitHub Copilot Premium Request Units (PRUs)' -publishDate: 2025-08-26 -params: - authors: - [ - { name: 'Dhruv Gupta', handle: 'dhruvg20' }, - { name: 'Greg Mohler', handle: 'CallMeGreg' }, - ] - -# Classifications of the framework to drive key concepts, design principles, and architectural best practices -pillars: - - cost-optimization - - operational-excellence - - security - - reliability - - performance-efficiency - -# The areas of the GitHub adoption journey. Inspiration taken from docs.github.com -areas: - - administration - - governance - - billing-management - - user-management - - monitoring-analytics - -# Classifications of industries who may be at different stages of the customer journey. -verticals: - - enterprise-software - - financial-services - - healthcare - - manufacturing - - technology - -# Individuals in key roles on the customer journey, typically consisting of one or more administrators and the end-user community. -personas: - - enterprise-administrator - - organization-owner - - billing-manager - - it-administrator - - development-manager - -# Deployment options for GitHub Enterprise, including Cloud (GHEC), Server (GHES), and Hybrid. -platform: - - github-enterprise-cloud - -# GitHub product functions designed to support every stage of development. -features: - - copilot-enterprise - - copilot-business - - premium-request-units - - copilot-chat - - coding-agents - - code-review-automation - -# Deeper-level topics of the GitHub Platform and its features. They are most often interacted with by end-users. -components: - - budget-management - - usage-monitoring - - license-assignment - - organizational-structure - - cost-controls - - reporting-analytics - -# Associated teams and other GitHub and Partner resources that can provide additional support. -github: - - customer-success - - enterprise-support - - professional-services - - technical-account-management ---- - - - - - - -## Scenario overview - -This playbook provides enterprise administrators with comprehensive guidance on managing GitHub Copilot Premium Request Units (PRUs). It includes operational procedures, cost management strategies, and administrative workflows essential for effective enterprise-scale deployment. - -Upon completion of this playbook, administrators will be able to: - -- Understand PRU fundamentals and billing mechanisms -- Configure enterprise-wide PRU budgets and policies -- Monitor and optimize PRU consumption across the organization -- Execute user license upgrades and organizational restructuring -- Implement cost controls and usage governance - -## Key design strategies and checklist - -### Pre-Implementation Assessment - -- [ ] Current user count and license distribution analyzed -- [ ] Projected PRU consumption calculated -- [ ] Budget approval obtained -- [ ] Organizational structure planned -- [ ] Administrative roles assigned - -### Configuration Phase - -- [ ] Enterprise PRU budget configured -- [ ] Organizational budgets established -- [ ] User licenses assigned appropriately -- [ ] Access controls implemented -- [ ] Monitoring and alerting configured - -### Post-Implementation Validation - -- [ ] User access to premium features verified -- [ ] Budget tracking systems operational -- [ ] Usage reporting mechanisms active -- [ ] Training materials distributed -- [ ] Support procedures communicated - -### Ongoing Operations - -- [ ] Monthly usage reports generated -- [ ] Budget performance reviewed -- [ ] Optimization opportunities identified -- [ ] Policy updates communicated -- [ ] User feedback collected and addressed - -## Assumptions and preconditions - -- **Budget Examples:** Dollar amounts ($100-$500) are realistic industry examples, not prescribed values -- **KPI Targets:** Percentages and thresholds are based on enterprise best practices, requiring organizational calibration -- **Timeframes:** Response times and implementation phases are suggested frameworks, not mandatory requirements -- **Team Structures:** Organizational models are generic examples requiring customization for specific enterprises -- **User Classifications:** Role-based categories are illustrative and should be adapted to actual organizational structures - -## Recommended deployment - -### Table of Contents - -1. [Fundamentals](#1-fundamentals) -2. [Planning & Configuration](#2-planning--configuration) -3. [Implementation Procedures](#3-implementation-procedures) -4. [Monitoring & Optimization](#4-monitoring--optimization) -5. [Administrative Reference](#5-administrative-reference) -6. [Troubleshooting](#6-troubleshooting) -7. [Appendices](#7-appendices) - -### 1. Fundamentals - -#### 1.1 Premium Request Units Overview - -**What are PRUs?** -Premium Request Units (PRUs) represent usage credits for advanced GitHub Copilot features that exceed standard plan allowances. These units enable access to: - -- Advanced AI models (Claude, Gemini, etc.) -- Agentic coding capabilities (Copilot cloud agent) -- Enhanced code review capabilities (Copilot Code Review) - -For governance-specific controls around cloud agents — policy configuration, security boundaries, MCP governance, and audit pipelines — see [Governing AI cloud agents in GitHub Enterprise]({{< relref "governing-agents.md" >}}). - -**Key Principles:** - -- PRUs reset monthly on the 1st at 00:00:00 UTC -- Unused PRUs do not roll over to the next month -- Default enterprise setting blocks all PRU overages ($0 budget) -- Administrative intervention required to enable premium features - -#### 1.2 Business Impact Assessment - -**Cost Implications:** - -- Base overage rate: $0.04 USD per premium request -- Potential for significant cost escalation without proper controls -- Direct impact on IT budget and resource allocation - -**Operational Benefits:** - -- Enhanced developer productivity with advanced AI models -- Improved code quality through advanced review features -- Accelerated development cycles with cloud agents - -### 2. Planning & Configuration - -Before enabling additional Premium Requests its important to understand which Copilot plan your users have been assigned and what each tier provides. Refer to the [GitHub Copilot plans documentation](https://docs.github.com/en/enterprise-cloud@latest/copilot/get-started/plans#comparing-copilot-plans) for more details. - -Also keep in mind that different models have different request multipliers, which changes the effective cost per request. Refer to the [GitHub Copilot model multipliers documentation](https://docs.github.com/en/enterprise-cloud@latest/copilot/concepts/billing/copilot-requests) for more details. - -A deliberate assignment strategy ensures: - -1. Power users (e.g., senior engineers, platform, AI productivity champions) aren’t throttled in their use of premium capabilities. -2. Broad populations (e.g., general developers or light consumers) aren’t over-provisioned. -3. Premium model usage (and therefore PRU spend exposure) stays concentrated where it produces more ROI. - -### 3. Implementation Procedures - -#### 3.1 Budget Configuration Workflows - -##### Scenario A: Enterprise-Wide PRU Enablement - -**Objective:** Enable PRU overages for all users across the enterprise - -**Prerequisites:** - -- Enterprise Owner or Billing Manager privileges -- Understanding of projected monthly PRU consumption -- Approved budget allocation - -**Procedure:** - -1. **Access Enterprise Settings** - - Navigate to GitHub Enterprise settings - - Select "Billing and licensing" → "Budgets and alerts" - -2. **Modify Default Budget** - - Locate current $0 budget setting - - Choose one of the following actions: - - **Option A:** Delete $0 budget (enables unlimited PRU usage) - - **Option B:** Increase budget to specific dollar amount - -3. **Validation** - - Verify budget change is reflected immediately - - Confirm all users can now access premium features - - Monitor initial usage patterns - -##### Scenario B: Selective User Enablement - -**Objective:** Enable PRUs for specific teams or user groups - -**Prerequisites:** - -- Organizational structure planning -- User classification and requirements analysis - -**Procedure:** - -1. **Create Organizational Boundaries** - - ```text - Enterprise → Organization A (Development Team) - → Organization B (QA Team) - → Organization C (Standard Users) - ``` - -2. **License Assignment** - - Assign Copilot Enterprise licenses to Organizations A & B - - Maintain Copilot Business licenses for Organization C - - Configure custom budgets per organization - -3. **User Migration** - - Move target users to appropriate organizations - - Validate license assignments - - Communicate changes to affected users - -##### Scenario C: Cost Center Management - -**Objective:** Implement granular budget controls with departmental allocation - -**Prerequisites:** - -- Organizational structure planning -- User classification and requirements analysis -- Cost center allocation - -{{< callout type="info" >}} -The following budget amounts and consumption estimates are provided as realistic examples for planning purposes. Organizations should adjust these values based on their specific requirements, team sizes, and usage patterns. -{{< /callout >}} - -**Example Implementation Framework:** - -| Organization/Cost Center | Monthly Budget | License Type | User Count | Estimated PRUs/Month | -| --------------------------- | --------------- | ------------ | ---------- | -------------------- | -| **Senior Development Team** | $500 | Enterprise | 15-20 | 300-500 | -| **QA/Testing Team** | $200 | Enterprise | 8-12 | 100-200 | -| **DevOps Team** | $300 | Enterprise | 5-8 | 200-300 | -| **Junior Developers** | $100 | Business | 20-30 | 50-100 | -| **General Users** | $0 (restricted) | Business | 100+ | 0 | - -**Consumption Patterns by Role:** - -- **Senior Developers:** Heavy usage of premium models (Claude Opus, o3) for complex problems -- **Junior Developers:** Moderate usage focusing on learning and code assistance -- **DevOps Engineers:** High usage for automation and infrastructure coding -- **QA Teams:** Targeted usage for test automation and code review - -{{< callout type="info" >}} -These budget amounts reflect typical enterprise deployments but should be validated through pilot programs and adjusted based on actual consumption patterns. -{{< /callout >}} - -#### 3.2 User License Management - -##### Upgrade Procedures - -##### Standard Business → Enterprise Upgrade - -**Step-by-Step Process:** - -1. **Assessment Phase** - - Identify users requiring premium features - - Calculate cost impact of upgrade - - Obtain budget approval - -2. **Organizational Restructuring** - - ```text - Current: Enterprise → Single Org → All Users (Business License) - - Target: Enterprise → Senior Dev Team (Enterprise + $500 budget) - → QA Team (Enterprise + $200 budget) - → Junior Devs (Business + $100 budget) - → General Users (Business + $0 budget) - ``` - -3. **Implementation Timeline** - - **Phase 1 (0-30 days):** Create organizations and assign core development team - - **Phase 2 (30-60 days):** Migrate QA and DevOps teams with budget allocation - - **Phase 3 (60-90 days):** Evaluate junior developer access and general user needs - - **Phase 4 (90+ days):** Optimize based on usage data and feedback - -4. **Verification** - - Test premium feature access - - Validate billing configuration - - Confirm user satisfaction - -### 4. Monitoring & Optimization - -#### 4.1 Usage Analytics & Reporting - -##### Monthly Reporting Checklist - -**Data Collection Points:** - -- [ ] Total PRU consumption by organization -- [ ] Top 10 users by PRU usage -- [ ] Feature-specific consumption patterns -- [ ] Model usage distribution -- [ ] Cost analysis and budget variance - -**Reporting Schedule:** - -- **Weekly:** High-level usage trends -- **Monthly:** Comprehensive cost analysis -- **Quarterly:** Strategic optimization review - -##### Key Performance Indicators (KPIs) - -{{< callout type="info" >}} -The following KPI targets are based on industry benchmarks and enterprise deployment patterns. Organizations should establish their own targets based on baseline measurements and business objectives. -{{< /callout >}} - -**Baseline Performance Targets:** - -| Metric Category | Target Range | Monitoring Focus | Optimization Trigger | -| ----------------------- | -------------------------- | ----------------------- | ---------------------- | -| **Monthly PRU Growth** | 10-20% month-over-month | Track adoption patterns | >25% unexpected growth | -| **Cost per Developer** | $10-30/month average | Budget efficiency | >$50/month per user | -| **Budget Utilization** | 70-85% of allocated budget | Spending control | >90% utilization | -| **Feature Adoption** | 60%+ active daily users | Usage effectiveness | <40% engagement | -| **Premium Model Usage** | 20-30% of total requests | Cost optimization | >50% premium usage | - -**Monthly Reporting Benchmarks:** - -- **Low Usage Organizations:** <100 PRUs/month per user -- **Medium Usage Organizations:** 100-300 PRUs/month per user -- **High Usage Organizations:** >300 PRUs/month per user - -**Cost Efficiency Targets:** - -- **Optimal Range:** 15-25% of total Copilot spend on PRUs -- **Alert Threshold:** >40% of budget on premium features -- **Review Trigger:** >60% budget utilization by mid-month - -{{< callout type="info" >}} -These targets should be calibrated based on your organization's development practices, team experience levels, and project complexity. -{{< /callout >}} - -#### 4.2 Cost Optimization Strategies - -##### Immediate Actions - -- **Model Usage Optimization:** Encourage use of included models (GPT-4.1, GPT-4o) for routine tasks -- **Feature Education:** Train users on cost-effective feature usage patterns -- **Budget Alerts:** Implement automated notifications based on organizational thresholds - -##### Medium-term Initiatives - -- **User Segmentation:** Classify users by actual vs. projected PRU consumption -- **License Optimization:** Right-size licenses based on usage patterns -- **Policy Development:** Establish governance guidelines for premium feature usage - -##### Long-term Strategic Initiatives - -- **Advanced Analytics:** Implement monitoring systems for PRU consumption patterns -- **Integration Assessment:** Evaluate third-party tool integrations for cost efficiency -- **Organizational Review:** Assess org structure effectiveness based on usage data - -#### 4.3 Operational Workflows - -##### Enterprise PRU Management Process Flow - -```mermaid -flowchart TD - A[Enterprise Default: $0 PRU Budget] --> B{Evaluate PRU Requirements} - B -->|All Users Need PRUs| C[Delete/Modify $0 Budget] - B -->|Selective Users Need PRUs| D[Create Targeted Organizations] - B -->|No PRUs Required| E[Maintain Current State] - - C --> F[Enterprise-wide PRU Access] - D --> G[Assign Users to Organizations] - G --> H[Apply Enterprise Licenses] - H --> I[Configure Custom Budgets] - - F --> J[Monitor Usage & Costs] - I --> J - E --> K[Periodic Review] - - J --> L{Budget Threshold Exceeded?} - L -->|Yes| M[Optimize Usage/Increase Budget] - L -->|No| N[Continue Monitoring] - - M --> J - N --> J -``` - -##### PRU Billing Hierarchy Structure - -The PRU Billing Hierarchy begins with the **Enterprise Admin**, who is responsible for setting enterprise-wide policies, managing billing, and monitoring overall usage. - -From the Enterprise Admin, three different types of organizations are defined: - -1. **Organization A (Team/Department)** - - Licenses are assigned based on team needs. - - Budget allocations are provided according to requirements. - - Feature access is configured as needed. - - Team members use resources based on their roles and have access to assigned models. - -2. **Organization B (Team/Department)** - - Similar to Organization A with licenses tailored to specific needs. - - Budget allocation and feature access are configured accordingly. - - Team members operate with role-based usage and access to models. - -3. **Organization C (Standard Users)** - - Designed for standard users with business licenses. - - Restricted or no PRU budget. - - Access limited to standard features. - - Standard users have basic Copilot features but no PRU access. - -**In summary:** -The Enterprise Admin governs and manages policies and budgets across all organizations, while each organization determines licenses and access based on business requirements. Team members and standard users then utilize features and resources aligned with their assigned roles and budget levels. - -{{< callout type="info" >}} -This diagram shows a generic organizational structure. Actual implementations should be customized based on specific organizational needs, team structures, and business requirements. -{{< /callout >}} - -### 5. Administrative Reference - -#### 5.1 Role-Based Access Control (RBAC) - -##### Administrative Roles and Permissions - -| Role Title | Scope | Core Permissions | PRU-Specific Capabilities | -| ---------------------- | --------------------- | -------------------------- | ------------------------------------------------------------------------------------------------------------------------- | -| **Enterprise Owner** | Enterprise-wide | Full administrative access | • Set enterprise PRU budgets
• Create/delete organizations
• View all usage reports
• Manage billing settings | -| **Organization Owner** | Organization-specific | Org-level administration | • Assign Copilot seats
• Manage org PRU budgets
• View org usage reports
• Configure member access | -| **Billing Manager** | Enterprise or Org | Financial management | • Monitor PRU costs
• Configure budget alerts
• Access billing reports
• Manage payment methods | -| **Repository Admin** | Repository-specific | Code management | • View repository usage
• Configure repository policies
• Access basic usage metrics | - -#### 5.2 Administrative Task Matrix - -##### Routine Operations (Daily/Weekly) - -| Task Category | Frequency | Responsible Role | Action Items | -| -------------------- | --------- | ------------------ | -------------------------------------------------------------------------------------------------- | -| **Usage Monitoring** | Daily | Billing Manager | • Review PRU consumption alerts
• Check budget utilization
• Validate unusual usage spikes | -| **User Support** | As needed | Organization Owner | • Handle access requests
• Resolve license assignment issues
• Provide feature guidance | -| **Cost Analysis** | Weekly | Billing Manager | • Generate usage reports
• Calculate cost per user
• Identify optimization opportunities | - -##### Strategic Operations (Monthly/Quarterly) - -| Task Category | Frequency | Responsible Role | Action Items | -| ------------------------ | --------- | ---------------- | ------------------------------------------------------------------------------------------------------- | -| **Budget Planning** | Monthly | Enterprise Owner | • Review budget performance
• Adjust allocations as needed
• Plan for next month's requirements | -| **License Optimization** | Quarterly | Enterprise Owner | • Analyze license utilization
• Right-size user assignments
• Evaluate plan upgrades/downgrades | -| **Policy Review** | Quarterly | Enterprise Owner | • Update usage policies
• Revise governance guidelines
• Communicate policy changes | - -#### 5.3 Configuration Management - -##### Budget Configuration Options - -| Configuration Type | Implementation Method | Use Case | Impact Level | -| --------------------------- | ---------------------------------------- | ----------------------------------- | ----------------- | -| **Global Unlimited** | Delete $0 enterprise budget | All users need premium access | High cost risk | -| **Global Limited** | Set enterprise budget to specific amount | Controlled enterprise-wide spending | Medium cost risk | -| **Organizational Budgets** | Create org-specific budgets | Departmental cost control | Low cost risk | -| **User-Level Restrictions** | Individual license management | Precise cost control | Minimal cost risk | - -##### Emergency Procedures - -{{< callout type="info" >}} -Response timeframes are suggested based on enterprise best practices. Organizations should adjust these based on their operational requirements and support capabilities. -{{< /callout >}} - -**Budget Overage Response Protocol:** - -1. **Immediate Response (0-2 hours)** - - Identify cause of overage (automated alerts recommended) - - Assess business impact and user disruption - - Implement temporary restrictions if necessary - -2. **Short-term Actions (2-24 hours)** - - Communicate with affected users and stakeholders - - Adjust budgets or implement policy changes - - Document incident details for analysis - -3. **Follow-up Analysis (24-72 hours)** - - Conduct root cause analysis with usage data - - Update policies and procedures based on findings - - Implement preventive measures and monitoring improvements - -**Escalation Response Times:** - -- **Critical Issues:** 1 hour response (service outages, security) -- **High Priority:** 4 hours response (budget exceeded >200%) -- **Medium Priority:** 24 hours response (policy violations, anomalies) -- **Low Priority:** 72 hours response (general questions, training) - -### 6. Troubleshooting - -#### 6.1 Common Issues and Resolutions - -##### Issue: Users Cannot Access Premium Features - -**Symptoms:** - -- Premium models not available in Copilot Chat -- cloud agents disabled or restricted -- Error messages about PRU limitations - -**Diagnostic Steps:** - -1. Verify user's organization assignment -2. Check organization's PRU budget status -3. Confirm license type (Business vs. Enterprise) -4. Review enterprise-level budget settings - -**Resolution Path:** - -```text -Problem Identified → Check Budget Status → Adjust Configuration → Validate Access → Monitor Usage -``` - -**Solutions:** - -- **Budget Issue:** Increase or delete $0 budget restriction -- **License Issue:** Upgrade user to appropriate license tier -- **Organizational Issue:** Move user to organization with PRU access - -##### Issue: Unexpected High PRU Consumption - -**Symptoms:** - -- Rapid budget depletion -- Usage alerts triggering frequently -- Costs exceeding projections - -**Investigation Protocol:** - -1. **Identify Top Consumers** - - Generate user-level usage reports - - Analyze feature-specific consumption - - Review model selection patterns - -2. **Pattern Analysis** - - Compare with historical usage - - Identify unusual activity spikes - - Correlate with business events - -3. **Optimization Actions** - - User education on cost-effective practices - - Policy adjustments for high-cost models - - Budget reallocation or restrictions - -#### 6.2 Escalation Procedures - -##### Severity Levels - -| Level | Criteria | Response Time | Escalation Path | -| ------------ | ------------------------------------ | ------------- | ---------------------------------- | -| **Critical** | Service outage, security breach | 1 hour | Enterprise Owner → GitHub Support | -| **High** | Budget exceeded by >50% | 4 hours | Billing Manager → Enterprise Owner | -| **Medium** | Usage anomalies, policy violations | 24 hours | Org Owner → Billing Manager | -| **Low** | General questions, training requests | 72 hours | User → Org Owner | - -#### 6.3 Preventive Measures - -##### Monitoring Alerts Configuration - -{{< callout type="info" >}} -These threshold percentages are based on enterprise financial management best practices. Organizations should adjust based on their risk tolerance and budget management processes. -{{< /callout >}} - -**Standard Alert Levels:** - -- **Early Warning:** 75% of monthly budget consumed -- **Critical Alert:** 90% of monthly budget consumed -- **Executive Notification:** 95% of monthly budget consumed -- **Automatic Restrictions:** 100% of monthly budget consumed - -**Usage Pattern Monitoring:** - -- **Anomaly Detection:** >3x average daily consumption -- **New User Monitoring:** Track first 30 days of usage patterns -- **Model Usage Alerts:** >50% premium model usage in department -- **Cost Efficiency Reviews:** Monthly cost-per-user above $50 - -**Recommended Monitoring Frequency:** - -- **Real-time:** Budget consumption and overage alerts -- **Daily:** Usage pattern analysis and anomaly detection -- **Weekly:** Cost efficiency and trend analysis -- **Monthly:** Comprehensive usage review and optimization planning - -{{< callout type="info" >}} -Alert thresholds should be calibrated during initial deployment and adjusted based on actual consumption patterns and business requirements. -{{< /callout >}} - -### 7. Appendices - -#### Appendix A: Quick Reference Guide - -##### Critical Facts Summary - -| Configuration Item | Default Value | Impact | Administrative Action Required | -| ------------------------- | -------------------------- | ------------------------- | --------------------------------- | -| **Enterprise PRU Budget** | $0 (overages blocked) | No premium feature access | Delete or modify budget to enable | -| **PRU Cost Rate** | $0.04 USD per request | Direct billing impact | Monitor and control via budgets | -| **Reset Schedule** | 1st of month, 00:00:00 UTC | Monthly consumption reset | Plan budgets accordingly | -| **Budget Change Timing** | Immediate effect | Real-time access control | Can adjust mid-month as needed | -| **Overage Capability** | Paid plans only | Feature availability | Free plans cannot exceed limits | - -## Seeking further assistance - -{{% seeking-further-assistance-details %}} - -## Related links - -- [GitHub Copilot Billing Documentation](https://docs.github.com/en/enterprise-cloud@latest/copilot/concepts/billing) -- [Managing Premium Request Allowance](https://docs.github.com/en/enterprise-cloud@latest/copilot/how-tos/manage-and-track-spending) -- [Enterprise Administration Best Practices](https://docs.github.com/en/enterprise-cloud@latest/admin) -- [GitHub API Documentation for Enterprise Management](https://docs.github.com/en/rest/enterprise-admin) -- [Managing premium request allowance for enterprise users](https://docs.github.com/en/enterprise-cloud@latest/copilot/how-tos/manage-and-track-spending/manage-for-your-enterprise) -- [GitHub Copilot requests overview](https://docs.github.com/en/enterprise-cloud@latest/copilot/concepts/billing/copilot-requests) -- [Copilot billing for organizations and enterprises](https://docs.github.com/en/enterprise-cloud@latest/copilot/concepts/billing/organizations-and-enterprises) diff --git a/content/library/governance/recommendations/governing-agents.md b/content/library/governance/recommendations/governing-agents.md index a8806ff..67db16c 100644 --- a/content/library/governance/recommendations/governing-agents.md +++ b/content/library/governance/recommendations/governing-agents.md @@ -370,7 +370,7 @@ Copilot agents consume both [GitHub Actions minutes and premium requests](https: - **Actions minutes add up.** Each agent session consumes GitHub Actions minutes for the duration of the agent's work. Monitor Actions usage to ensure agent workloads do not impact your CI/CD capacity. - **Model selection amplifies cost.** Different models have different [request multipliers](https://docs.github.com/en/copilot/concepts/billing/copilot-requests). Factor model choice into your cost governance. -For detailed operational procedures — budget configuration workflows, license tier upgrades, cost center allocation, KPI benchmarks, and troubleshooting — see [Managing GitHub Copilot Premium Request Units (PRUs)]({{< relref "copilot-policies-best-practices/copilot_pru_enterprise_admin_playbook.md" >}}). +For detailed guidance on budget configuration, cost center allocation, user-level budgets, and cost attribution — see [Managing AI credits with FinOps principles]({{< relref "managing-ai-credits.md" >}}). {{< callout type="info" >}} Cost governance is part of AI governance. Predictable costs keep adoption sustainable. Revisit budgets quarterly as you scale usage or adopt new features. @@ -413,7 +413,7 @@ Review quality depends heavily on custom instructions. Without instructions, rev - [GitHub Enterprise Policies & Best Practices]({{< relref "governance-policies-best-practices" >}}) — general platform security hardening checklist (rulesets, CODEOWNERS, commit signing, audit log streaming) - [Rulesets Best Practices]({{< relref "rulesets-best-practices.md" >}}) — guidance on push vs. branch rulesets, file-path scoping, and CODEOWNERS integration - [Application Security Checklist]({{< relref "/library/application-security/checklist" >}}) — foundational code security practices (CI checks, pull review, secret scanning) that agent governance builds on -- [Managing GitHub Copilot Premium Request Units (PRUs)]({{< relref "copilot-policies-best-practices/copilot_pru_enterprise_admin_playbook.md" >}}) — budget configuration, license tier management, KPI benchmarks, and cost optimization procedures +- [Managing AI credits with FinOps principles]({{< relref "managing-ai-credits.md" >}}) — layered budget configuration, user-level budgets, cost attribution, and usage governance ### External resources diff --git a/content/library/governance/recommendations/managing-ai-credits.md b/content/library/governance/recommendations/managing-ai-credits.md new file mode 100644 index 0000000..1b816a9 --- /dev/null +++ b/content/library/governance/recommendations/managing-ai-credits.md @@ -0,0 +1,615 @@ +--- +# SPDX-FileCopyrightText: GitHub and The Project Authors +# SPDX-License-Identifier: MIT + +draft: false # Set to false when ready to publish + +title: 'Managing AI credits' +publishDate: 2026-05-13 +params: + authors: + [ + { name: 'Kitty Chiu', handle: 'kittychiu' }, + ] + +# Classifications of the framework to drive key concepts, design principles, and architectural best practices +pillars: + - governance + - productivity + - cost-optimization + - operational-excellence + - security + - reliability + - performance-efficiency + +# The areas of the GitHub adoption journey +areas: + - enterprise-and-teams + - ci-cd-and-devops + - administration + - governance + - billing-management + - user-management + - monitoring-analytics + +# Classifications of industries who may be at different stages of the customer journey +verticals: + - finance + - information-technology + - government + - manufacturing + - professional-service + - retail + - enterprise-software + - financial-services + - healthcare + - technology + +# Individuals in key roles on the customer journey +personas: + - administrator + - project-manager + - enterprise-administrator + - organization-owner + - billing-manager + - it-administrator + - development-manager + +# Deployment options for GitHub Enterprise +platform: + - github-enterprise-cloud + - github-enterprise-cloud-plus-emu + +# GitHub product functions designed to support every stage of development +features: + - copilot + - copilot-enterprise + - copilot-business + - copilot-ai-credits + - copilot-chat + - coding-agents + - code-review-automation + +# Deeper-level topics of the GitHub Platform and its features +components: + - billing-and-cost-centers + - insights-and-metrics + - governance-and-policy + - budget-management + - usage-monitoring + - license-assignment + - organizational-structure + - cost-controls + - reporting-analytics + +# Associated teams and other GitHub and Partner resources +github: + - customer-success-architect + - customer-success-manager + - enterprise-support + - expert-services + - microsoft + - customer-success + - professional-services + - technical-account-management +--- + + + + +## Scenario overview + +Managing AI credits requires governance practices similar to those used for cloud +infrastructure: pooled consumption, variable demand, tiered budget controls, +and cost attribution. It also requires right-sizing model selection, managing +context window usage, and accounting for task complexity. + +Each Copilot license contributes a monthly AI credit allowance to a shared +enterprise pool, and usage beyond that pool requires additional spend. Usage is +driven by individual behavior rather than automated workloads, and +the shared pool model means one user's consumption directly affects another's +available included AI credits in the same enterprise account. + +This article provides opinionated recommendations for managing AI credits and usage-based billing (UBB) through the lens of the +[FinOps Framework](https://learn.microsoft.com/en-us/cloud-computing/finops/framework/finops-framework). +It defers to the authoritative sources for GitHub billing budget configuration +and focuses instead on design decisions, trade-offs, and governance strategies. + +{{< callout type="info" >}} +This article assumes the post-1 June 2026 AI credits model rollout unless otherwise noted. +{{< /callout >}} + +## Key design strategies and checklist + +### 1: Use layered budgets as guardrails, not as allocation + +Design budgets as concurrent guardrails, not as a top-down allocation tree. +The layered budgets (enterprise-level → cost center → user) enforce spending +limits at multiple levels simultaneously. The most restrictive budget applies: +a user can be blocked by any budget layer when enforcement is enabled, even if their ULB +and the enterprise-level budget both have remaining balances. + +This approach prevents runaway spend while preserving user productivity. +Each layer serves a different governance purpose: enterprise-level budgets cap total +organizational exposure, cost center budgets delegate accountability to cost owners, +and user-level budgets prevent individual overconsumption of shared resources. + +User-level budgets (ULBs) are recommended as the only hard enforcement caps, while enterprise and cost center budgets act as monitoring guardrails unless explicitly configured to stop usage. + +### 2: Separate included AI credits governance from additional spend governance + +Design distinct governance approaches for each spending phase. A billing cycle +has two phases: the included AI credits phase (where usage draws from the pre-paid +pool) and the additional spend phase (where usage incurs additional charges +after the pool is exhausted). For the included AI credits phase, use ULBs to +prevent any single user from exhausting the shared pool before others benefit. +For the additional spend phase (post-pool spend), use enterprise and cost +center budgets to cap the additional cost your organization is willing to incur. + +All budget layers apply at all times; this separation keeps your governance model coherent. ULBs are your tool for +fair consumption of the shared pool. Enterprise and cost center budgets are +your tool for controlling incremental cost exposure. Mixing their purposes — +for example, relying only on the enterprise-level budget while ignoring ULBs — leaves +the pooled AI credits ungoverned and vulnerable to a few heavy users +draining it before month-end. + +### 3: Establish cost attribution before you optimize + +Build a reporting pipeline that answers "Which cost centers consume what, and at what +cost?" before making any budget or policy changes. GitHub provides a +downloadable CSV usage report with per-user, per-model, per-day breakdowns. +Use this as the foundation for all optimization decisions. + +Without attribution, budget adjustments and model restrictions become guesswork. +Cost visibility is a prerequisite for informed governance — you cannot +right-size budgets or identify waste without first measuring consumption +patterns. + +### 4: Align usage guidance with cost transparency + +Publish guidance that ties model tier to task complexity and makes the cost +difference visible to users. Token costs vary significantly across +models, and context window usage can drive up consumption quickly. + +When users understand the cost implications of model selection and context window usage, they can +make informed trade-offs between capability and consumption. This is the AI +equivalent of right-sizing compute instances: match the resource to the task. + +### FinOps checklist + +The Start phase +establishes minimum viable governance, the Adopt phase adds cost center-level +controls and reporting, and the Advance phase automates and scales. + +#### Start phase + +- Design the layered budget structure with stakeholder input +- Define roles and assign response ownership for usage threshold alerts +- Communicate budget policies and thresholds to all Copilot users before + enforcement begins +- Set an enterprise-level AI credit budget as the baseline monitoring guardrail for additional + spend +- Set a universal ULB to prevent any single user from consuming a + disproportionate share of the pooled AI credits +- Identify power users (e.g. agentic workflows, frontier model usage) and plan + custom ULB overrides + +#### Adopt phase + +- Create cost centers mapped to business units and assign additional spend + budgets to each +- Apply custom ULB overrides for defined profiles (e.g. power users) +- Build a cost attribution and reporting pipeline using the AI + usage CSV report +- Map usage data to users and cost centers for showback reporting +- Publish usage guidance that ties model tier to task complexity +- Establish the operating model with alert response process and task cadence +- Establish a review cadence: weekly during rollout, monthly once stable + +#### Advance phase + +- Automate budget management via the REST API (bulk ULB overrides, cost + center mapping, budget adjustments) +- Implement chargeback +- Forecast pooled AI credits exhaustion based on observed trends +- Integrate AI credit reporting with enterprise FinOps tooling + +## Assumptions and preconditions + +- Your organization uses **GitHub Enterprise Cloud** (GHEC) with Copilot + Business or Copilot Enterprise licenses. +- You have **enterprise owner** or **billing manager** access to configure + budgets and cost centers. +- Your enterprise has an established FinOps practice, or at minimum a role + responsible for metered cost governance. +- Cost centers in GitHub billing are already configured, or you are willing to + set them up to align with your monitoring requirements. +- You have access to the enterprise AI usage report (CSV download available + from billing settings). +- User identities can be mapped to business units for cost attribution. + +## Recommended implementations + +### Step 1: Design the layered budget structure + +GitHub's AI credit budgeting uses a layered structure with three budget levels: enterprise-level, cost center, and ULBs. Each layer serves a different governance purpose, and they operate concurrently to enforce spending limits. See [Understanding Copilot budgeting](https://support.github.com/product-guides/github-copilot/get-started/understanding-copilot-budgeting) for definitions and how these budget layers interact. + +Design your budget layers by asking targeted questions at each governance level: + +**1. User-level: What is the responsible monthly spend per user?** + +- What was the average individual spend during the pilot? +- Where are we in the AI SDLC rollout plan? +- What new agentic use cases are we adopting in the coming months? +- What new AI capabilities are we building? +- What is the persona and usage-tier mix (developer, reviewer, SRE, data, business user)? + +**2. Scoped level: Which teams or groups need their own additional spend budget on top of the pooled AI credits?** + +- What is each group's forecast consumption above the pooled allowance? +- Which projects depend on frontier models or agentic workflows that consume credits faster? +- How variable is their month-to-month usage (seasonality, release cycles)? +- How do we treat shared platform teams vs product teams in the allocation (showback vs chargeback)? +- Who is the accountable budget owner authorized to lift or hold a scope-level cap? +- Which time-bound budgets must be ring-fenced (POCs, hackathons)? +- What signal tells us a scope cap is too tight (throttling productive work) vs too loose (no friction on waste)? + +**3. Enterprise-level: How much autonomy should cost owners have over spending?** + +- What is the total enterprise AI envelope for the fiscal year? +- What is our monetary commitment vs on-demand PAYG strategy? +- What share sits in the pooled credit reserve vs pre-allocated to cost centers? +- What guardrails, alert thresholds, and approval gates apply to additional spend? +- How do we handle chargeback or showback across business units once pooled credits are exhausted? +- What is the escalation path when the pool is depleted mid-cycle (throttle, top-up, or block)? +- How do we forecast for headcount changes, M&A, and new AI capability launches? + +### Step 2: Communicate budget accountability to stakeholders + +Before configuring anything, ensure all stakeholders — users, cost +owners, and finance — understand how the layered budgets affect them, +where accountability lies, and how they can monitor their own usage. + +#### Assess your change management needs + +For users and cost owners new to layered AI credit budgets, publishing diagrams and example scenarios can help them understand budget accountability and what happens when limits are reached. + +#### Key points to communicate + +- **Each user is accountable for their own budget.** The ULB is the ultimate + spend cap per user. Once exhausted, the user is blocked from all Copilot + features (except code completion and next edit suggestions) regardless of remaining budget at any + other layer. +- **Cost center budgets govern scoped additional spend.** Users assigned + to a cost center share that additional spend budget. The + cost owner is accountable for monitoring and managing that budget. +- **Users not assigned to a cost center fall under the enterprise-level budget.** + If a user is not mapped to any cost center, their additional spend is + governed solely by the enterprise-level budget. Ensure these users are + identified and accounted for. +- **The most restrictive budget applies.** Whichever budget has the least + remaining amount determines whether a request proceeds — a user can be + blocked by any layer. + +Note that default user-facing messages may not clearly indicate which budget layer caused a block. Support processes and self-serve dashboards should account for this. + +#### Example scenarios + +Publish concrete example scenarios like the following to make the budget rules tangible for users: + +Example scenario 1: +> **"I have budget left, why am I blocked?"** +> Alex has a ULB of $200. The enterprise included AI credits pool is still +> available, and Alex's cost center has $10K remaining. But Alex has already +> consumed $200 in AI credits from the shared pool this month. Alex is +> blocked because the ULB is the per-user hard cap — the remaining balance +> in other budgets does not override it. + +Example scenario 2: +> **"The enterprise budget is exhausted — will the platform stop?"** +> The Platform Engineering cost center has a $50K additional spend budget. +> The enterprise-level budget ran out on the 20th, and cost center usage is configured to **exclude** from the enterprise budget. +> The Platform Engineering team can continue using Copilot because their cost center budget still has remaining balance. +> Only users not assigned to Platform Engineering cost center are affected by the exhausted enterprise budget. + +#### Enable self-serve usage monitoring + +Provide users and cost owners with a way to check their own consumption so they can +self-govern usage before hitting their ULB and cost center budgets. For example, users can [view usage in their IDE](https://docs.github.com/en/copilot/how-tos/manage-and-track-spending/monitor-premium-requests#viewing-usage-in-your-ide). + +### Step 3: Set the enterprise-level additional spend budget + +Navigate to your enterprise billing settings and create an AI credit budget +sized to your total enterprise AI envelope for the fiscal year, adjusted for +your monetary commitment and PAYG arrangement. +This is your **baseline** guardrail for additional spend. Set this even if you +plan to use cost center budgets, because users not assigned to any cost center are +only governed by this budget. + +#### Enterprise-level budget configuration + +- Budget type: Bundled AI credit budget +- Scope: Enterprise +- Option: Exclude cost center usage +- Disable: "Stop usage when budget limit is reached" +- Enable: Budget threshold alerts (75%, 90%, 100%) +- Assign alert recipients: billing manager(s) and enterprise owner(s) + +Recommend **disabling "Stop usage when budget limit is reached"** at the enterprise level to avoid a single global cap that causes engineering disruptions. + +#### Enterprise-level budget scope option + +If you are planning to implement cost center budgets, decide +whether the enterprise-level budget should **include** or **exclude** cost center +usage: + +- **Include** (default): Cost center usage counts against both the cost center + budget and the enterprise-level budget. The enterprise-level budget acts as a global cap. +- **Exclude**: Cost center usage is only governed by the cost center budget. + Users in cost centers can continue spending even after the enterprise-level budget + is exhausted. Users outside cost centers are still governed by the enterprise + budget. + +Recommend **exclude** for most enterprises, especially if a significant +share of spend is pre-allocated to cost centers rather than held in a shared +reserve. With this option, each cost center's additional spend budget acts +as a _dedicated and isolated_ budget for that group — +if other cost centers exhaust the enterprise-level budget, cost centers with remaining +balance continue operating without interruption. This prevents one cost center's +overconsumption from blocking others' productivity. It also +simplifies budget forecasting: each cost owner can plan against their +own budget ceiling without accounting for others' spending patterns. See [Centralized enterprise budget vs. delegated cost center budgets](#centralized-enterprise-budget-vs-delegated-cost-center-budgets). + +### Step 4: Create budgets with cost centers + +For Copilot AI credits in a GHEC enterprise, cost center budgets must be attached to +a **cost center**. Create one or more cost centers and assign an AI credit +budget to each, sized to each scope's forecast consumption above the pooled +AI credits, with additional allowance for month-to-month variability. This answers the question: "When the pooled AI credits run +dry, how much additional spend is _this_ group allowed to incur?" + +{{< callout type="info" >}} +A user can only belong to **one cost center**. However, **multiple budgets** +still apply simultaneously (ULB, cost center budget, and enterprise-level budget). +{{< /callout >}} + +When creating a cost center, you need to choose either +**organization** or **user-based** scope. See [Management methods](https://docs.github.com/en/enterprise-cloud@latest/copilot/concepts/billing/premium-request-management). + +Recommend **user-based cost centers** when your organizational +structure does not map cleanly to budget ownership, or when users frequently +span multiple organizations. User-based cost centers give you direct control +over exactly which users fall under each budget without requiring changes to +org membership or Copilot license assignment. For enterprises with clean +org-to-team mapping and SCIM provisioning, organization-based cost centers +reduce ongoing maintenance. + +{{< callout type="info" >}} +**ULB and user-based cost center budgets are two different layers**. +ULBs are _user-level_ spend caps that apply to each user regardless of cost center assignment. +User-based cost centers are _groups of users that share_ a cost center budget for additional spend. Users in a user-based cost center are still subject to their ULB. +{{< /callout >}} + +Examples of how to scope user-based cost centers, informed by responses from +[Step 1](#step-1-design-the-layered-budget-structure): + +- **By role or function:** group all platform engineers into one cost center + and application developers into another, regardless of which orgs they + belong to. This addresses how to treat shared platform teams vs product + teams in the allocation. Roles with higher AI intensity (e.g. agents, + frontier model usage) get budgets sized to their forecast consumption + above the pooled allowance, while lighter roles operate under tighter + guardrails. +- **By budget tier:** create a "power users" cost center for staff whose + projects depend on frontier models or agentic workflows, and a "standard + users" cost center for everyone else. Size each tier's budget to its + actual consumption profile rather than applying a uniform cap that either + over-provisions or under-serves. +- **By business unit:** assign users from, for example, Finance, Engineering, + and Product each to their own cost center to enable chargeback or showback + aligned with P&L ownership. Designate an accountable budget owner per + cost center who is authorized to lift or hold the scope-level cap. +- **By project or initiative:** temporarily group users working on a + high-priority initiative (POCs, hackathons, time-bound projects) into a + dedicated cost center with a ring-fenced additional spend budget. This + lets you measure the AI credit cost per deliverable and evaluate ROI. + +If there is planned month-to-month +variability (seasonality, release cycles), proactively monitor whether the cap is +throttling productive work or providing too little friction on waste. + +{{< callout type="info" >}} +If you already use cost centers for chargeback across other products +(Actions, Code Scanning), you cannot create a second set of cost centers +purely for AI credit budgeting. Use ULBs for user-level differentiation +in this scenario. +{{< /callout >}} + +#### Scoped budget configuration + +- Budget type: Bundled AI credit budget +- Scope: Cost center (organization-based or user-based) +- Disable: "Stop usage when budget limit is reached" +- Enable: Budget threshold alerts (75%, 90%, 100%) +- Assign alert recipients: billing manager(s) and cost owner(s) + +### Step 5: Set user-level budgets + +User-level budgets (ULBs) define the _individual spending authority_ for each +user — the total amount of AI credits a single user can consume in a billing +period, regardless of whether those credits come from the shared pool or +additional spend. ULBs are the only budget layer that **caps total user consumption across both pooled credits and additional spend**, making them critical for preventing a small number +of heavy users from exhausting the shared pool before others can benefit. + +#### Recommended approach + +1. Set a universal ULB as the default for all users. Size it using average + individual spend from your pilot, plus additional margin for your current rollout + stage and planned AI capability growth. + See [ULBs and developer productivity](#ulbs-and-developer-productivity). + +2. Identify power users early — those adopting agentic use cases, frontier + models, or new AI capabilities — and set custom ULB overrides sized + to their persona and usage pattern. Each custom ULB is a 1:1 + assignment to a user. + +3. To unblock a user mid-month, apply custom ULB overrides temporarily + until the next billing cycle. + +#### Universal ULB configuration + +- Budget type: Bundled AI credit budget +- Scope: Users +- Enable: "Stop usage when budget limit is reached" + +### Step 6: Build a cost attribution and reporting pipeline + +GitHub provides a downloadable CSV usage report. +This data includes `aic_quantity` (AI credits consumed) and +`aic_gross_amount` (estimated USD cost). Build a reporting pipeline that maps usage to users and cost centers, and surface insights to inform budget adjustments and model guidance. See [Prepare for usage based billing](https://docs.github.com/en/copilot/how-tos/manage-and-track-spending/prepare-for-usage-based-billing). + +#### Cost attribution approaches + +- **Chargeback model:** At month-end, use the usage report to identify + consumption by user. If one business unit's users consumed a + disproportionate share of the shared pool, allocate those costs back to that + unit using your internal finance processes. +- **Showback model:** Publish per-cost center dashboards without charging back. This + creates accountability and awareness without the overhead of internal + billing. + +### Step 7: Publish usage guidance + +Token costs vary dramatically by model, and context window size drives +consumption further. Provide users with guidance on: + +- **Responsible context usage:** Minimize unnecessary context in instructions to + reduce token consumption. +- **Model selection by task complexity:** Match model tier to the task. See + [models comparison](https://docs.github.com/en/copilot/reference/ai-models/model-comparison) + and + [models pricing](https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing). + +### Step 8: Establish the operating model + +Define ownership and the tasks required to manage AI spend. + +#### Roles and responsibilities + +| Role | Scope | AI credit responsibilities | +|---|---|---| +| Enterprise Owner | Enterprise and Org | Set policies, configure budgets and alerts, view all usage reports, forecast overall AI credit needs | +| Billing Manager | Enterprise and Org | Configure budgets and alerts, generate usage reports, monitor consumption, adjust budgets | +| Cost Owner | Cost center | Monitor scope usage, respond to alerts, right-size own budget | + +#### Task cadence + +| Task | Frequency | Owner | +|---|---|---| +| Review consumption alerts | Daily | Billing Manager | +| Generate usage reports and cost-per-user analysis | Weekly | Billing Manager | +| Review budget performance and adjust allocations | Monthly | Cost Owner | +| Right-size budgets | Quarterly | Cost Owner | +| Update governance guidelines | Quarterly | Enterprise Owner | + +#### Response process for budget threshold alerts + +Define who +receives each alert and what action they take: + +- **75%:** Billing Manager reviews consumption trajectory. No action required + unless the trend projects exhaustion before month-end. +- **90%:** Billing Manager escalates to cost owner. Evaluate whether to + increase the budget or restrict usage. +- **100%:** If "Stop usage" is enabled, users are blocked automatically. + Cost Owner decides whether to raise the budget or leave the block in + place until the next billing cycle. + +### Step 9: Review and adjust on a monthly cycle + +UBB consumption patterns will emerge over time. During the first three +months, review weekly. After that, shift to monthly reviews aligned with your +billing cycle. + +Review checklist: + +1. Compare actual consumption against budgets at enterprise, cost center, and user levels. +2. Identify users or cost centers consistently hitting budget limits, interview them to understand usage practices and emerging patterns, and detect anomalies (e.g. unattended or autonomous usage). +3. Analyze usage trends, particularly when during the month the pooled AI credits are exhausted. + Consider whether to right-size additional spend budgets, add ULB + overrides, or publish guidance to educate responsible usage. +4. Aggregate weekly data to identify the top 5–10% of consumers. +5. Remediate and update forecasts based on observed trends. + +## Additional solution details and design considerations + +### Profile usage based on consumption patterns and use cases + +Not all users consume AI credits equally. Some may be power users running agentic workflows with frontier models, while others may be light users leveraging smaller models for code completion. Profiling users based on their consumption patterns and use cases can help you plan and forecast more accurately, and identify who may need custom ULB overrides. + +Examples of usage patterns: + +- Piloting new AI capabilities and enablement +- Agentic workflows with frontier models +- Migration factory with Copilot agents +- Landing zones operations with Copilot agents +- Burst usage in POCs, hackathons, tech conferences, and time-bound projects +- Doubling down on seasonal projects (e.g. retail app for Black Friday, end-of-quarter functions, tax season) + +Proactively profile users and their usage patterns, configure cost center budgets, and assign custom ULB overrides. This prevents users from being blocked mid-task while still maintaining overall cost governance. + +### ULBs and developer productivity + +The goal of a ULB is to manage unplanned spend and cap outlier consumption, +not to restrict typical usage. + +Initially, set a high universal ULB to avoid blocking productive work. If the universal ULB is too low, it risks blocking users mid-task. Blocked users lose access to all Copilot features (except code completion and next edit suggestions) until the next billing cycle, or until an admin intervenes with custom ULB overrides. + +Start with the per-seat included AI credits value plus a margin. You can use +pilot data to inform this margin. Monitor for the first two billing cycles, +then periodically review and adjust as user behavior and use cases evolve. + +### Centralized enterprise budget vs. delegated cost center budgets + +| Approach | Benefit | Risk | +|---|---|---| +| Single enterprise-level budget, no cost centers | Simple to manage | No cost center-level accountability; hard to attribute costs | +| Cost centers excluded from enterprise-level budget | Each cost center has a dedicated and isolated budget | Total enterprise spend is delegated; requires individual cost center budget sizing | +| Cost centers included in enterprise-level budget | Global cap plus cost center visibility | Enterprise-level budget can block others that still have cost center budget remaining | + +For organizations with 100+ seats, use cost centers mapped +to organizations with the enterprise-level budget set to **exclude** cost center +usage. This gives each cost center a _dedicated_ amount that +cannot be consumed by other cost centers — cost owners can forecast against their own +budget without other cost center dependencies. The enterprise-level budget then acts solely +as a backstop for users not assigned to any cost center. Set a universal +ULB as the cross-cutting safety net. + +### Scaling budget management for large enterprises + +For enterprises with thousands of users, managing custom ULBs through the +UI is impractical. Use the [GitHub REST API](https://docs.github.com/en/enterprise-cloud@latest/rest/billing/budgets?apiVersion=2026-03-10) to: + +- Bulk-create custom ULB overrides. +- Programmatically map users to cost centers. +- Set and adjust enterprise-level and cost center budgets. + +## Seeking further assistance + +{{% seeking-further-assistance-details %}} + +## Related links + +{{% related-links-github-docs %}} + +### External resources + +- [Understanding Copilot budgeting](https://support.github.com/product-guides/github-copilot/get-started/understanding-copilot-budgeting) +- [FinOps Framework](https://learn.microsoft.com/en-us/cloud-computing/finops/framework/finops-framework) +- [Usage management methods in enterprise](https://docs.github.com/en/enterprise-cloud@latest/copilot/concepts/billing/premium-request-management) +- [Monitoring your GitHub Copilot usage and entitlements](https://docs.github.com/en/copilot/how-tos/manage-and-track-spending/monitor-premium-requests) +- [Prepare for usage-based billing](https://docs.github.com/en/copilot/how-tos/manage-and-track-spending/prepare-for-usage-based-billing) +- [Viewing your usage of metered products and licenses](https://docs.github.com/en/billing/how-tos/products/view-productlicense-use) +- [Models comparison](https://docs.github.com/en/copilot/reference/ai-models/model-comparison) +- [Models and pricing for GitHub Copilot](https://docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing) +- [GitHub REST API: Billing budgets](https://docs.github.com/en/enterprise-cloud@latest/rest/billing/budgets?apiVersion=2026-03-10) diff --git a/content/library/overview/release-notes/2025-q3.md b/content/library/overview/release-notes/2025-q3.md index 0d949c0..9664ce6 100644 --- a/content/library/overview/release-notes/2025-q3.md +++ b/content/library/overview/release-notes/2025-q3.md @@ -4,6 +4,6 @@ - **Update: [Repository Management Enhancement](/library/governance/recommendations/managing-repositories-at-scale/)** - Updated the "Managing repositories at scale" article with opinionated guidance on adopting rulesets and custom properties to meet business objectives, including actionable strategies for governance at scale - **Update: [GitHub Actions Policy Updates](/library/application-security/recommendations/actions-security/)** - Updated the GitHub Actions recommendations with new policy capabilities and more prescriptive governance and security guidance for managing workflows at scale -- **New Content: [GitHub Copilot Enterprise Administration](/library/governance/recommendations/copilot-policies-best-practices/copilot_pru_enterprise_admin_playbook/)** - Published an enterprise playbook for managing GitHub Copilot Premium Request Units (PRUs), including budget configuration, KPI targets, monitoring, and cost control strategies +- **New Content: GitHub PRU Enterprise Admin Playbook** - Published an enterprise playbook for managing GitHub Copilot Premium Request Units (PRUs), including budget configuration, KPI targets, monitoring, and cost control strategies. UPDATED: Deprecated on May 2026. - **New Content: [Security Alert Management](/library/application-security/recommendations/prioritizing-alerts/)** - Published a scenario for prioritizing security alert remediation using GitHub's built-in metadata and organizational context, including practical guidance on implementing GitHub's security campaigns and vulnerability triage workflows - **New Content: [Champion Program](/library/collaboration/recommendations/champion-program/)** - Published a recommendation for champion programs that empower engaged employees to guide peers through AI-driven change. diff --git a/docs/taxonomies.md b/docs/taxonomies.md index 29a5f34..43f73f5 100644 --- a/docs/taxonomies.md +++ b/docs/taxonomies.md @@ -71,6 +71,7 @@ areas: - collaborative-coding - community - context-engineering + - continuous-delivery - developers - enterprise-and-teams - getting-started @@ -141,7 +142,6 @@ components: - third-party-agents - code-review-automation - code-scanning - - context-engineering - copilot-autofix - copilot-business - copilot-chat