When the Agent Works, the Bill Grows: Build the AI Cost Model Before You Scale

Anthropic's IPO filing contains a striking line. The value of the company's own internal Claude usage, at list prices, "has grown by over 250 times from January 2026 to September 2026". In April, that monthly value "first exceeded our monthly personnel expense."

Read it carefully. This is usage valued at the prices customers pay, not Anthropic's actual compute bill, which the filing doesn't disclose. But the lesson holds for everyone else: when AI works, people use it more, and the bill follows.

A pilot's cost doesn't predict production

A pilot runs with ten users and a few hundred tasks a month. Take an illustrative agent that costs $0.40 per completed task. In the pilot, that is $80 a month, an easy line to approve. Roll the same agent out to the whole team at 40,000 tasks a month and the same price becomes $16,000.

And that assumes cost per task stays flat. It usually doesn't, because four things grow after launch:

• Volume: more users, more tasks and more edge cases than the pilot ever saw.
• Context: longer histories and bigger documents make every call more expensive.
• Retries: failed tool calls and loops are billed like successful ones.
• Review: the people checking the output are part of the cost too.

Five lines to write before you scale

• Volume: tasks per month at full rollout, not in the pilot.
• Cost per completed task: every model call, retry and tool call, divided by tasks that actually finished.
• Human time: minutes of review and approval per task.
• Ceiling and alert: a monthly cap, and who is told when spend reaches 80% of it.
• Owner: one person who reads the numbers every month.

Three numbers to read every month

Once the agent is live, watch the trend in cost per completed task as volume grows; compare it with what the same task costs when a person does it; and check where spend will land against the ceiling before the month ends. None of this is complicated. Most pilots simply never write it down.

How we approach it at Zylen

Zylen is an AI optimization partner. We work inside the AI systems companies already run and improve cost, security and data control one measured change at a time. Our ten-day AI Optimization Audit builds the cost model for your agents: what each task costs, where it will grow, and what to fix first.

Know the bill before it grows: start the AI Optimization Audit →

Source: Anthropic IPO filing, pp. 187–188, as reported by Business Insider, November 2026. Example figures are illustrative.