# Ten agents in production, none crossing a department line

> IDC found half of companies run 10+ agents but only 3% scale them across departments. Counting agents measures activity, not an operation.

- Type: Opinion
- Published: 2026-11-05
- Author: Sentynel
- URL: https://sentynel.ai/sygnals/ten-agents-none-crossing-departments/
- Language: en

A company with ten agents in production sounds advanced. An IDC study for AWS suggests it is common, and also suggests it is not the milestone it appears to be: **the number of agents is the easy thing to grow, and it says almost nothing about whether the operation changed.**

## What the study found

AWS's page for the IDC study reports that 50% of organizations had 10 or more agents in production in 2025, and that "just 3% of organizations are scaling agentic AI across departments." It also says that fewer than seven percent are in full production with at least one use case. The study surveyed more than 900 organizations across 15 industries and 10 countries.

Read together, those numbers describe the same company: plenty of agents, each one living inside a single team.

**Agents in production vs agents scaling across departments**

- 50%: have 10 or more agents in production
- 3%: are scaling agents across departments

Source: IDC study for AWS, as reported on aws.amazon.com/isv/resources/agentic-ai-idc-study. Adoption figures refer to 2025.

## Why agents stay inside one department

An agent is easy to place inside one team. The team owns the inbox, the data, the approver and the definition of "done". Nothing has to be agreed with anyone else.

Crossing a department line is a different kind of work. A request that starts in sales and ends in finance needs a shared record, a rule for who owns it at each moment, and an agreed meaning for every status. None of that is a model problem. **It is the part of the operation nobody wrote down**, so ten agents can run well and still hand the work back to people at every border.

This is why a tenth agent adds less than the first. The first one removes manual work. The tenth one adds another place where a handoff can fail.

## The border is where the operation lives

Most of the time a business loses is not spent doing a task. It is spent between tasks: a customer waits for a reply that is waiting for an approval that is waiting for someone to notice it. An agent that is excellent inside its box does nothing about that wait.

So the useful question is not "how many agents do we have?" It is "where does work cross a line between two teams, and what travels with it?"

**What has to cross a department line**

1. One record — The same case, visible to both teams
2. An owner — Named at every moment
3. Shared statuses — One meaning of done
4. An exit path — What happens when it fails

## What it looks like when the path is visible

Sentynel's Rodo case is a small example of building the path first. The retailer's WhatsApp, social and web enquiries were brought into one service operation with a measurable resolution path. In the first measured window, 82% of live-created enquiries were resolved. Sentynel's case record states the limits: the cohort covers 31 August to 24 September 2026, and the figures describe pipeline records, not all conversations.

The point is not the percentage. It is that once every enquiry has a place and a path, you can say what happened to it. That is what an agent needs before it can work across channels, and it is what a stack of department-level agents does not give you.

## A test for this week

Pick the last piece of work that crossed from one team to another, and trace it. Write down who held it at each step, what they could see, and how long it waited. If you cannot answer in an hour, you do not have an agent problem. You have an operation to design, and the agents come after it.

If you can answer, count how many of your agents touched that path. The honest number is usually lower than ten.

## Sources

- https://aws.amazon.com/isv/resources/agentic-ai-idc-study/
