AI Agents
The real cost of an agent in production
Model tokens are the visible cost. Retrieval, observability, and human triage are the ones that surprise you.
Published by CloudFixies · 02 December 2025
The line items
Model calls, embeddings, vector search, log storage, monitoring, escalation triage. Model tokens are usually 40% of the total.
The realistic budget
Assume £0.10–£0.40 per resolved user request for a well-tuned agent on a real workload. Cheaper is a red flag — you probably aren’t retaining logs.
◆ Takeaway
Budget for the whole pipeline, not just the model bill.