GVGanivra

Workflow Rolling out Claude Code or Codex

Roll out AI coding agents to every team, and know what it costs.

Going from a pilot to the whole engineering org raises the same questions every month: who is using it, which seats sit idle, whether each person is on the right plan, and what each team spends. Ganivra answers them from real usage.

  • Adoption week by week, against the seats you pay for
  • Idle seats and seat fit for every developer
  • Spend and budgets by team, project and client
Start with your pilot Estimate the rollout

14-day free trial, no credit card · From $29/month after the trial

Seats & adoption · Week 6Example data
Seats in use34 / 40
Idle seats6
Idle a month$750
9Wk 118Wk 224Wk 329Wk 433Wk 534Wk 640 seats
  • Heavy7
  • Good fit19
  • Light8
  • No usage6
Illustrative figures to show the layout, not real usage or typical results.

At a glance

Rolling out AI coding agents with Ganivra, in short.

Updated

Who it's for
Engineering, platform and finance leaders rolling out Claude Code or OpenAI Codex to many developers.
Measures
Active developers by week, seats in use and idle, seat fit, and cost by developer, team, project and model.
Deploy
Claude Code managed settings or device management for every machine, or one command per machine on macOS and Linux.
Not covered
Chat usage in ChatGPT Enterprise or Claude for Work, and Cursor. GitHub Copilot only as general model calls.
Data sent
Token counts, model, time, developer and labels. Never prompts, responses or source code.

Pilot to steady state

The questions change as the rollout grows.

Each stage has its own questions. The same data answers all of them, so you set it up once, at the start of the pilot.

  1. 1

    Pilot

    10 to 20 developers, a few weeks
    • Is it worth expanding?
    • What would the whole team cost?
    • Which teams get the most out of it?

    Ganivra showsCost per developer and per active day from real usage, which models they use, and a projection to your full headcount.

  2. 2

    Rollout

    Team by team
    • Who has started using it?
    • Are we inside each team's budget?
    • Is every machine sending data?

    Ganivra showsActive developers week by week against seats, spend by team with budget alerts, and developers using a tool without a seat.

  3. 3

    Steady state

    Every month
    • Which seats sit idle?
    • Is each person on the right plan?
    • What do we charge back?

    Ganivra showsIdle seats and their monthly cost, seat fit for every developer, a chargeback CSV by team or project, and re-pricing on cheaper models.

Before you expand

What would the whole team cost?

Enter what your pilot spent at API rates. Ganivra shows that figure for a real pilot; until then, the starting value uses Anthropic's published average.

Per developer a month$260
After rollout, a month$52,000
A year$624,000

The starting pilot spend is Anthropic's reported Claude Code average of about $13 per developer per active day, over 20 working days. Replace it with your pilot's real figure from Ganivra. The projection assumes the wider team works like the pilot; pilots are often the keenest users, so treat it as an upper estimate and check it after the first month.

Seats & adoption

Every seat checked against real usage.

Paste your seat list from the Claude or ChatGPT admin console: email, tool, plan and monthly price. Ganivra compares each seat with what that developer used.

No usage

No requests in the period. Reassign or cancel the seat.

Light

Usage worth under half the seat price. Per-token billing or a smaller plan may cost less.

Good fit

Usage in line with what the seat costs.

Heavy

Usage worth over twice the seat. Good value; check they aren't hitting plan limits.

Usage is valued at API list rates and scaled to a month. Light and heavy are rules of thumb, under half and over twice the seat price, not what your plan bills. Developers using a tool without a seat on the list are shown too, so the list stays complete.

Deploy

Set it up once for every machine.

Claude Code's managed settings send usage from every machine, and developers can't turn them off. Smaller teams can run one command per machine instead.

  • Claude Code, managedA managed-settings.json on each machine, a Jamf, Kandji or Intune profile, or the Claude Team or Enterprise admin console. Managed settings guide →
  • CodexDistribute ~/.codex/config.toml with your device-management tool, or use the setup command.
  • Teams and projectsCommit a .claude/settings.json to each repository with its team, project and client. Labels guide →
Or one command per Mac or Linux machine
curl -fsSL https://ganivra.com/install.sh | sh

managed-settings.json · macOS: /Library/Application Support/ClaudeCode/ · Linux: /etc/claude-code/

{
  "env": {
    "CLAUDE_CODE_ENABLE_TELEMETRY": "1",
    "OTEL_LOGS_EXPORTER": "otlp",
    "OTEL_METRICS_EXPORTER": "none",
    "OTEL_EXPORTER_OTLP_PROTOCOL": "http/protobuf",
    "OTEL_EXPORTER_OTLP_ENDPOINT": "https://api.ganivra.com/v1/otlp",
    "OTEL_EXPORTER_OTLP_HEADERS": "Authorization=Bearer gv_live_...",
    "OTEL_LOGS_EXPORT_INTERVAL": "60000",
    "OTEL_RESOURCE_ATTRIBUTES": "team.id=platform"
  }
}

Coverage

What it measures, and what it doesn't.

Measured

  • Claude Code: terminal, IDE and desktop app
  • OpenAI Codex CLI and the ChatGPT desktop app
  • AI features you build, through the SDK or OpenTelemetry

Not measured

  • Chat in ChatGPT Enterprise or Claude for Work
  • Cursor, for now
  • GitHub Copilot in the coding-agents view (its spans arrive as model calls)

Questions

AI coding rollout FAQ

Comparing tools first? Try the AI coding tools cost calculator, or read about charging costs back to teams.

How do we measure an AI coding pilot?

Connect the pilot developers' Claude Code or Codex to Ganivra before the pilot starts. You get cost per developer and per active day, the models they use and the projects they work on, from real usage. Add the pilot's seats to see whether each one is used, then project the per-developer cost to your full team.

How much does Claude Code cost per developer?

Anthropic reports an average of about $13 per developer per active day on API billing, with 90% of users under $30 a day. Your team's figure depends on how much they use agents and which models they choose, so measure it in a pilot rather than budgeting from the average.

How do we find unused Claude or ChatGPT seats?

Paste your seat list into Ganivra's Seats & adoption view: each developer's email, the tool, the plan and the monthly price. Ganivra matches it with recorded usage and lists seats with no requests in the period, with what they cost each month.

How do we roll out the settings to every developer?

For Claude Code, put the export settings in managed settings: a file or device-management profile on each machine, or the Claude Team or Enterprise admin console. Developers can't override them. For Codex, distribute ~/.codex/config.toml with your device-management tool, or have each developer run the one-command setup.

Does it track ChatGPT Enterprise or Claude chat usage?

No. Ganivra measures Claude Code and OpenAI Codex, which export per-request telemetry, plus AI features you build yourself. Chat usage in ChatGPT Enterprise or Claude for Work isn't exported per request, so it isn't covered. Cursor isn't supported yet.

Can we set a budget for each team?

Yes. Label each developer's team, then create a budget for that team in Ganivra with warning and critical thresholds. Spend is checked as usage arrives, and the dashboard shows each team's spend against its budget.

Measure the pilot before you roll it out.

Connect the pilot developers, add their seats, and see real cost per developer before you commit the whole team. 14-day free trial, no credit card, from $29 a month after.

Start free trial See cost tracking
Then on each pilot machine
curl -fsSL https://ganivra.com/install.sh | sh