Inside Terrain: How 17 AI Agents Analyze Your Cloud Costs

Terrain does not use a single AI model to analyze your cloud costs. It uses 17 specialized agents, each designed for a specific aspect of cost intelligence. Think of it as a team of analysts, each with deep expertise in one domain, working in parallel on your data.

Here is what each agent does and how they work together.

The Strategic Layer

finops_chief is the orchestrator. When you ask a question or start a conversation, the finops_chief determines which agents to activate, synthesizes their findings, and delivers a unified response. It ensures that every analysis is grounded in FOCUS-compliant data and aligned with FinOps best practices.

report_writer transforms technical analysis into executive-ready summaries. When you need to present findings to leadership, the report writer generates narratives that connect cost data to business impact -- no spreadsheets required.

The Detection Layer

anomaly_detective runs a 3-model ML ensemble (KNN, Isolation Forest, Local Outlier Factor) to identify spending patterns that deviate from normal. A 40% spike in compute costs last Tuesday? The anomaly detective catches it, determines its significance, and flags it with a confidence score.

waste_detective_agent hunts for idle resources, zombie instances, and unattached volumes -- the infrastructure equivalent of leaving the lights on in an empty building. It identifies resources that cost money but deliver no value.

governance_agent monitors tagging compliance and policy violations. Untagged resources are invisible to chargeback systems. The governance agent finds them and flags the gap.

The Optimization Layer

efficiency_agent recommends rightsizing and auto-scaling adjustments. That m5.2xlarge running at 15% CPU utilization? The efficiency agent calculates the savings from downsizing and provides the implementation script.

rate_optimizer_agent analyzes your usage patterns against Reserved Instance, Savings Plan, and Spot Instance pricing. It calculates the optimal commitment level for each service and region.

cost_optimizer implements multi-strategy optimization by combining insights from the efficiency agent, rate optimizer, and waste detective into a prioritized action plan.

ai_cost_optimizer is purpose-built for AI spend. It analyzes token usage across models and providers, identifies opportunities for model right-sizing, prompt optimization, and provider routing.

The Analysis Layer

allocation_agent handles tag-based chargeback and showback. It maps costs to teams, projects, and business units using your tagging taxonomy.

tagging_agent analyzes tag coverage and consistency across your cloud estate. Missing tags mean unallocated costs. Inconsistent tags mean inaccurate allocation.

forecast_analyst_agent uses Prophet and StatsForecast to predict future spend based on historical patterns. It accounts for seasonality, growth trends, and one-time events.

focus_insights_agent normalizes multi-cloud data using the FOCUS specification, enabling apples-to-apples comparison across AWS, Azure, and GCP.

sustainability_agent estimates the carbon footprint of your cloud infrastructure and identifies opportunities to reduce environmental impact alongside cost.

historical_trend_agent provides long-term trend analysis, identifying gradual shifts in spending patterns that might not trigger anomaly alerts but represent significant cost trajectory changes.

The Onboarding Layer

day1_intelligence_agent activates during your first conversation with Terrain. It analyzes your initial data to surface immediate insights -- quick wins, obvious waste, and baseline metrics -- so you get value from day one.

qa_agent validates the outputs of other agents, checking for data quality issues, calculation accuracy, and consistency. It is the quality assurance layer that ensures every recommendation is trustworthy.

How They Work Together

When you ask "Why did costs spike last week?", here is what happens:

1. The finops_chief routes the question to the relevant agents
2. The anomaly_detective identifies the spike and its characteristics
3. The waste_detective and efficiency_agent check for contributing factors
4. The allocation_agent attributes the spike to specific teams and services
5. The forecast_analyst determines if this is a trend or a one-time event
6. The qa_agent validates all findings
7. The report_writer synthesizes everything into a clear answer

The entire process takes 30 seconds. The same analysis done manually takes 2-4 hours.

That is the difference between intelligence and dashboards. Dashboards show you data. Intelligence tells you what to do about it.