The result we can share
During a two-week trial with a stealth company, Open45 recovered $5,400. For an early engagement, that is a concrete result around which to have a useful conversation about agent costs.
The customer’s identity and the tasks involved remain confidential. This case study therefore focuses on the reported outcome and the principles behind our work, rather than publishing a detailed account of the customer’s systems.
Why we focus on the cost of useful work
An agent’s total bill is only one part of the picture. More spend can reflect more useful work. It can also reflect repeated context, unnecessary retries, or a mismatch between a task and the resources used to complete it.
Our approach is to investigate that distinction and connect the findings to practical next steps. We use Claude Opus 5.5 within our analysis workflow, while keeping customer-specific material and internal implementation details private.
Those are general principles of the Open45 approach. We are not disclosing which patterns, tools, or changes were involved in this customer’s trial.
Putting the result in context
The reported recovery is $5,400 over the two-week trial. The customer’s baseline spend and detailed breakdown remain private, so we present the dollar result without a percentage-reduction claim or annual projection.
This is an early, founder-reported result rather than an independently audited study. A different workload will have a different opportunity, and future results require their own evaluation.
Our focus is the economics of useful work: understand the cost, identify a change, and establish whether it helps.
Start with your own workload
If you’re running agents in production and the bill is getting harder to explain, a focused spend review is a practical place to start. We’ll discuss your setup, the evidence available, and what a useful outcome would look like.