Your Agent Bills While It Waits. Here's the Fix.
"Most of an agent's life is spent waiting — on a tool, a human, the next step — and the whole time you're holding a live process awake and billing for it." — Hamza Tahir, ZenML, AI Engineer 2026
That single sentence, dropped in an AI Engineer conference talk this week, names the unit-economics problem that no agent demo video ever mentions. The problem that decides which agent companies survive contact with a CFO.
Every agent product demo shows the same clip: the agent plans, calls tools, produces output — a continuous stream of productive work. What the demo never shows is what happens between those steps. The tool call that takes 8 seconds to return. The human-approval gate where someone is in a meeting. The retry backoff after a rate limit. The polling loop waiting for a webhook. The overnight batch job that parks a process for 6 hours.
That is where your money goes. Not inference. Not model selection. The waiting.
