We Cut Our AI Agent API Spend by 79% — Here's the Exact Fix

We were spending $198/month on AI agent API costs. Then we figured out where the money was actually going. It wasn't the LLM calls we knew about. It was a loop we'd forgotten to audit. One agent loop was running every 60 seconds and calling 4 external tools per cycle. That's 5,760 tool calls per day. Multiplied by a month: 172,800 tool calls nobody was watching. The Root Problem: No Cost Attribution When you run multiple AI agents, API costs pool together. Without attribution, you can't answer basic questions: Which agent is burning the most tokens? Which tool calls are expensive vs. cheap? Which loop cadence is appropriate vs. excessive? We had no answers to any of these. So we built a cost attribution pattern into every agent loop. The Fix (3 Steps) Step 1: Tag every LLM call with an agent ID Every API call gets metadata: { "agent": "suki", "loop": "content-loop", "task": "draft-tweet" } . Log this to a daily cost file. Step 2: Set loop cadence intentionally Don't run a loop every 60