Economics
The real cost is the second application
Everyone budgets for the first governed workflow. Almost nobody budgets for the tenth. That's where the math actually shows up.
How the first one gets paid for
The first agent workflow gets attention because it's visible. There's a sponsor, a date, and enough urgency to cover whatever it takes. A custom integration, a couple of specialist contractors, a few months of engineering time.
It ships. Everyone's happy. Somebody writes it up as a win.
Then a second team wants to deploy an agent, and asks the obvious question: what's this one going to cost?
Two ways that question can go
Scroll horizontally to see all columns.
| Priced per application | Priced for a portfolio | |
|---|---|---|
| First application | A funded integration project | A funded integration project |
| Second application | Another integration project | Configuration against policy you already have |
| Policy | Written again for this workflow | Already published and versioned |
| Field names and values | Negotiated again between teams | Already in the catalog for that industry |
| Review routing | Built again | Already defined by role |
| Evidence and export | Built again | The same records your auditors have seen before |
| Where it tops out | However many projects you can fund | However many applications you have |
What the left-hand column does to a program
It caps it. Not through any decision anyone made out loud. It's that every new workflow has to win its own budget fight against the cost of a fresh integration, and most of them lose.
You can spot this from across the room. One beautifully instrumented pilot, and a slowly growing number of agents running nowhere near it. Nobody chose that outcome. It's what happens when application two costs about what application one did.
Gartner's forecast that more than 40% of agentic AI projects will be canceled by the end of 2027 lists three causes, and the first one is escalating costs. Uncertain business value and inadequate risk controls follow. Those are not independent problems. A program that has to fund a fresh integration per workflow will struggle to show value across a portfolio, because it never gets a portfolio.
Some of the industry's pilot-to-production problem is technical, and some of it is organizational. A decent chunk of it is arithmetic.
For context on the size of the line item: IDC put security and governance for AI and agents at 16.7% of planned AI investment worldwide. That is a real budget, and how far it goes depends almost entirely on whether it buys reusable infrastructure or a series of one-off projects.
What has to carry over
Flattening that curve isn't a pricing trick. It comes down to whether four things survive from the first workflow to the next one.
Policy that outlives one workflow
Rules published centrally and versioned, so the second application points at the same policy instead of restating it in fresh code.
A vocabulary you didn't invent
Field names and permitted values pulled from the standards your industry already uses, so team two isn't relitigating what to call a payment status.
Review routing tied to roles
Who's allowed to decide an exception is a fact about your organization, not about the workflow. Set it once and the next application inherits it.
One evidence pipeline
Same record shape, same export, same verification. An auditor who's read one of your applications already knows how to read the next.
When those hold, adding an application is configuration. When they don't, it's a project, and your program stops at however many projects finance will approve.
Run the math with your own numbers
Put in what an implementation costs you per application, your workload, and your labor rates. The arithmetic argues better than we can.