The Missing Line Item

AI Radar is Jasmin Guthmann’s monthly column on Agentic AI, business transformation, and the future of digital value creation. With many years of experience in technology, transformation, and AI, she is one of the established voices shaping the conversation around the future of digital business. In April 2025, her book If There Is a Will, There Is a Way was published — a no-compromise guide to unlocking potential by focusing on what truly matters. Her message: take action, embrace responsibility, and stand for what you believe in.

In the diva-e Conclusion column AI Radar, she shares insights from her practical experience, provides context on current developments, challenges trends, and explores what technological change means for businesses, leadership, and organizations.

An RFP landed on my desk a few weeks ago. Serious company, serious budget, serious intent. The core question, spelled out in a spreadsheet: how many FTEs can agents replace in year one? Savings modeled to the second decimal. And the cost of running those agents? Not in the document. No line for tokens. Nothing for orchestration, monitoring, evaluation, or the human who reviews the output when the agent is confidently wrong. The savings had a spreadsheet. The costs had a hope. 

That RFP is not an outlier. It is the current state of enterprise AI math. 

Here is the part the spreadsheet missed: an agent costs money every time it thinks. Every retry, every tool call, every context window it drags along. Software used to be a license fee and done. Agents are a workforce with a fuel bill. The marginal cost of hiring one is near zero. The marginal cost of it working is very real, and it scales with exactly the volume you were hoping to automate. 

Which means the honest question is not “how many FTEs does this replace?” It is: what does one resolved ticket cost when an agent handles it end to end, failures included? One processed claim. One qualified lead. Cost per outcome. Some agentic workflows are brutally cheap at that level. Some are more expensive than the humans they replace. They just fail faster and in nicer prose. You only know which one you own if you measure it. 

Almost nobody measures it. 

Last month I wrote that the emperor is naked. A lot of you nodded. Quick September update: he still hasn’t seen the invoice for the wardrobe. 

Because this is the same disease as pilot purgatory, one column further right in the budget. Pilots fail to scale because nobody built the architecture. Business cases fail to survive contact with production because nobody ran the unit economics. Both come from the same reflex: deciding on slides, in FTEs and vibes, instead of in outcomes and euros. 

The objection I hear at this point: “Token prices keep falling, this will solve itself.” Prices per token fall, yes. Consumption per task explodes. Agents that plan, retry, call tools, and check their own work burn multiples of what a single prompt did in 2024. Betting your business case on the price curve outrunning your consumption curve is not a strategy. It is a lottery ticket with a burn rate. 

Easy for a columnist to point at other people’s spreadsheets. Fair. So here is our side of it. 

The cost blindness is a symptom. The disease sits deeper. You cannot price what you cannot see, and most companies cannot see what their agents touch, what they do, or what they return, because nothing underneath connects them. That is the architecture problem from last month’s column, showing up in the CFO’s inbox. 

For the past months we have been building our answer to that deeper problem. I am not naming it yet. Something big is coming, and it will be unveiled at Conclusion Connect 2026 on September 17 in Düsseldorf. 

I will be on stage moderating the day. My team will be there. So will clients who run this in production and can tell you, invoice in hand, what agents actually cost and actually return. No slideware answers. Ask them the questions your RFP forgot. Bring the spreadsheet. We will fill in the missing line.