Why This Matters

Without budgets, AI costs spiral and latency frustrates. Setting both deliberately keeps Copilot productive without breaking the bank or the developer experience.

Concept Explained

Cost budgets: per-developer monthly credit allowance. Latency SLOs: max acceptable response time per surface. Fallback: cheaper model or no-AI when over budget or SLO violated.

How It Works

Estimate per-dev monthly cost (Pro ~$10, Pro+ ~$39, plus top-ups). Set latency SLOs (inline <500ms, chat <5s first token). Track via Metrics API. Fallback to cheaper model or disable Copilot when over.

Step-by-Step Tutorial

1. Estimate per-dev cost

2. Set latency SLOs

3. Track burn

4. Plan fallback

5. Adjust

Real-World Example

A team set $30/dev/month budget. Tracked weekly. Found 20% of devs exceeded; they were heavy coding-agent users. Switched them to cheaper models for routine tasks, kept agent for high-value chores. All devs back under budget within a month.

Example Prompts / Commands / Code

Surface              Cost/dev/month   Latency SLO
--------------------------------------------------
Inline completions    $5               <500ms
Chat                  $10              <5s first token
CLI                   $5               <3s first token
Coding agent          $10              <10min per task
Extensions            $5               <5s
--------------------------------------------------
Total                 $35/dev/month

Fallback:
- Over budget: switch to GPT-5 mini for routine tasks.
- SLO violated: disable Copilot temporarily; investigate.

# weekly-cost-report.py
import requests
import os

# Pull Copilot usage via REST API
org = 'your-org'
token = os.environ['GITHUB_TOKEN']
r = requests.get(f'https://api.github.com/orgs/{org}/copilot/usage',
                 headers={'Authorization': f'Bearer {token}'})
usage = r.json()

for user in usage['users']:
    if user['credits_used'] > 35:  # over budget
        print(f'{user["login"]}: ${user["credits_used"]} - OVER BUDGET')

Common Mistakes

• No budgets — costs spiral.
• No latency SLOs — Copilot feels slow, devs abandon.
• No fallback — over budget means hard cut.
• Not tracking weekly — surprises at month-end.

Best Practices

• Set per-dev monthly cost budget.
• Define latency SLOs per surface.
• Track weekly via Metrics API.
• Plan fallback (cheaper model, disable).
• Review quarterly; adjust.

Troubleshooting

Practical Exercise

Professional Challenge

Frequently Asked Questions

Further Reading

• Metrics API

Official References

• Metrics API

Related lessons: PR-01, PR-03