All agents

Watches each customer’s list of suppliers and sends an alert when a supplier’s score drops, a payment remark appears or the board changes, with a short note on what changed.

Runs per day

5

p95 run time

5.7 s

Tokens per day

19,496

Failed runs

2.9 %

From the sample telemetry, 2026-09-18 – 2026-09-24, which includes the incident on 2026-09-22.

Facts

service.name

safegrade-supplier-monitor

Model

claude-haiku-4-5

Runs

Every 4 hours

Tools

load_portfoliodetect_changeswrite_alertsend_alert

Outcome to optimize

Alert precision: At least 90 % of alerts marked useful by customers (safegrade.alert.sent)

Human oversight

Customers rate each alert; low-rated alert types are reviewed weekly.

EU AI Act

Informs a customer’s decision. Relies on Scorekeeper, a high-risk system.

Reads from

Scorekeeper, Debt Watch, Board Screener

Read by

Customers directly

Auga lenses

Cost Optimization, Operate

What can go wrong

• Symptom: Customers mute alerts after a burst of false alarms

  Root cause: Alerts fired on score drops that came from stale financial figures

Telemetry

Each run is one trace. Besides the standard gen_ai.* attributes, the root span carries:

Shape of one run:

{
  "name": "invoke_agent Supplier Monitor",
  "resource": {
    "service.name": "safegrade-supplier-monitor",
    "service.namespace": "safegrade"
  },
  "attributes": {
    "gen_ai.operation.name": "invoke_agent",
    "gen_ai.provider.name": "anthropic",
    "gen_ai.agent.id": "supplier-monitor",
    "gen_ai.agent.name": "Supplier Monitor",
    "gen_ai.request.model": "claude-haiku-4-5",
    "safegrade.portfolio.id": "…",
    "safegrade.portfolio.suppliers": "…",
    "safegrade.alert.sent": "…"
  },
  "children": [
    "chat claude-haiku-4-5",
    "execute_tool load_portfolio",
    "execute_tool detect_changes",
    "execute_tool write_alert",
    "execute_tool send_alert",
    "chat claude-haiku-4-5"
  ]
}

Download the full week (OTLP/JSON)