Managing AI Agent Costs Across a Dev Team

When one developer uses AI agents, cost tracking is simple: check your billing dashboard. When a team of three or more developers each uses different agents on different projects, cost visibility becomes a real management problem. This article walks through how Styrby's Power tier handles team cost management.

The Scenario

Consider a team of three developers working on a SaaS product:

• Developer A uses Claude Code (Opus 4) for backend architecture and database work. Heavy usage, averaging $30/day.
• Developer B uses Claude Code (Sonnet 4) and Codex for frontend work. Moderate usage, averaging $12/day.
• Developer C uses Gemini CLI for research and Aider for legacy code. Light usage, averaging $6/day.

Total team spend: roughly $48/day, or about $1,000/month. At this scale, cost management matters, but without aggregated data it is not obvious where the money is going.

Shared Dashboards

The Power tier includes a team dashboard that aggregates costs across all team members. The team lead sees:

• Total team spend by day, week, and month
• Per-developer breakdown
• Per-agent breakdown (how much is going to Anthropic vs. OpenAI vs. Google)
• Per-project breakdown (which projects consume the most AI budget)

Individual developers see their own costs and the team total. They do not see other developers' individual costs unless the team lead enables full visibility. This is a deliberate choice: cost tracking should inform, not surveil.

Per-Developer Attribution

Every session in Styrby is tied to a user account. When Developer A starts a Claude Code session, the costs are attributed to their account. The team dashboard rolls these up into per-developer totals.

This matters for two reasons:

1. Budget allocation. A team lead can set different daily budgets for different developers based on their role and typical usage. Developer A doing architecture work with Opus needs a higher budget than Developer C doing research with Gemini.
2. Cost optimization. If one developer consistently spends 3x more than peers on similar tasks, that is a signal to investigate. They may be using a more expensive model than necessary, or their prompts are triggering retry loops.

Team Budget Alerts

Budget alerts work at both the individual and team level:

# Team-level budget (team lead sets this)
styrby team budget set --period daily --limit 75 \
  --notify-at 80 --stop-at 100

# Individual budgets (per developer)
styrby team budget set --user dev-a@team.com \
  --period daily --limit 40

styrby team budget set --user dev-b@team.com \
  --period daily --limit 20

styrby team budget set --user dev-c@team.com \
  --period daily --limit 15

When Developer A hits their $40 daily limit, they get a notification and their sessions pause. The team budget continues to track against $75. If Developer B has an unusually expensive day and pushes the team total past $75, the team-level alert fires for the team lead.

Model Selection Guidance

The team dashboard highlights model usage patterns. A common finding: developers default to whatever model they set up initially and never change it. Developer A might use Opus 4 for everything, including tasks where Sonnet 4 would produce identical results at one-fifth the cost.

The dashboard shows cost per model and per task type. If 60% of Developer A's Opus sessions are short tasks (under 20 turns), that is a signal that many of those sessions could use Sonnet instead. The team lead can share this data without mandating model choices.

Project Cost Attribution

When multiple developers work on the same project, their costs are aggregated under the project tag. This answers questions like "how much AI budget did the checkout-redesign project consume?" across all developers and agents.

Project attribution requires that developers start sessions with a project tag, which happens automatically when the CLI detects a git repository:

# Project is auto-detected from the git remote
styrby connect --agent claude
# → Project: checkout-redesign (from git remote)

# Or specify manually
styrby connect --agent claude --project checkout-redesign

Monthly Cost Reports

The Power tier generates monthly cost reports that can be exported as CSV or JSON. These reports include:

• Total team spend by agent and model
• Per-developer totals
• Per-project totals
• Budget alert history (how many alerts fired, how many were overridden)
• Model efficiency metrics (cost per session, cost per turn)

For companies that need to justify AI tool spend to management, these reports provide the data. "We spent $1,100 on AI agents this month across 3 developers and 4 projects" is a more useful conversation starter than "our Anthropic bill was $800."

What Power Tier Costs

The Power tier is $49/month for the account, supporting up to 3 team members. For a three-person team, that is $49/month for cost visibility into $1,000/month of AI spend. The math works if the visibility helps you save more than $49 through better model selection, retry loop detection, and budget enforcement. For most teams at this spend level, it does.