- Stacked subscriptions (ChatGPT + Claude + Copilot)$600
- Max-plan premium over mid-tier$1,920
- Wrong-model usage (premium for mini-class tasks)$4,800
- Idle / forgotten API budget$2,400
- Failed re-runs & retries$1,800
- Duplicate-tool overlap$2,880
What an AI-native engineer wastes in a year.
Three ways your AI spend is lying to you.
Model Over-qualification
Running GPT-4 on tasks that GPT-3.5 handles identically. The most common source of invisible waste.
Redundant Calls
Identical prompts fired multiple times per session. Cache-friendly patterns left on the table.
Token Bloat
System prompts averaging 4,000 tokens when 800 would produce identical output quality.
Combined: $14,400/yr per engineer × your team size = _____
Your AI P&L, updated in real time.
npm install. Three lines. Done.
CLI
$ npm install -g @getverbal/cli
$ verbal init
$ verbal status
✓ Connected to Verbal
✓ Tracking 3 API keys
✓ First report in ~24 hoursMCP CONFIG
{
"mcpServers": {
"verbal": {
"command": "verbal",
"args": ["mcp-serve"],
"env": {
"VERBAL_API_KEY": "vbl_…"
}
}
}
}OpenAIAnthropicAWS BedrockAzureGoogle AIOllamaLiteLLM
“AI spend is infrastructure spend. It deserves the same discipline, the same visibility, and the same tooling that we built for compute, storage, and network over the last twenty years. The teams that treat it this way will outperform the ones that don’t.”
Patrick Hill
14 years at Atlassian · Built Compass · Co-authored the Incident Management Handbook