Think of everything in your business that does work without a person pressing a button. The text that goes out after a missed call. The review request after a job is closed. The website chat. The connection someone set up two years ago that copies new leads into a spreadsheet.
Now try to list them all, with who set each one up and how to turn it off. Most owners get partway and stop. That gap is worth closing before you add anything new.
A big-company problem that starts small
On September 24, at its Succeed conference, the data company Dataiku launched a product called Agent Management. Its whole job is to find the AI agents a company is running across the platforms it uses and put them on one list, with an owner and a cost for each. Dataiku says it will be generally available in October.
Dataiku’s CEO, Florian Douetteau, explained the problem this way: “Ask a bank how many servers it runs, and you get an answer to the decimal. Ask how many AI agents it’s running, and you get a shrug or a guess.” In the company’s announcement he added a harder line: “almost nobody can fire an agent.”
The product is built for large companies. The problem is not. A service business with a dozen people can have the same shrug, just on a smaller scale.
How a shop ends up with automations nobody owns
Here is a made-up example. An HVAC company’s office manager sets up an automation that texts customers a review request when a job is marked done. It works well. A year later she moves on.
Months after that, the team starts marking jobs done from the truck, before the tech has packed up. Customers now get a “how did we do?” text while the tech is still in their driveway. Nobody knows where the automation lives. It runs under a login that belonged to the person who left. It takes an afternoon to find and switch off.
Nothing in that story is malicious. Automations are easy to set up now, and each one makes sense on the day it is built. Trouble starts later, when nobody is sure who owns which one, what it can touch, or how to stop it.
The one-page list
You do not need special software to fix this. A spreadsheet with one row per automation is enough. Fill in every column. If you cannot, that row is the one to look at first.
- Name and job. One sentence. “Texts a review request when a job is closed.”
- Owner. A person’s name, not “the office.”
- Where it lives. Which app, and which login it runs under. A login that belongs to someone who has left is a warning sign.
- What it can touch. Customers, your inbox, your calendar, invoices, payments.
- What it can do. Read only, or can it send messages, change records, take payments or delete things? Anything that sends to customers or moves money deserves a closer look.
- Monthly cost. The app, the AI usage and any extra seats it needs.
- How to turn it off. The exact steps, and who is allowed to do it.
- Next check date. Every 90 days the owner confirms it is still needed. If nobody can say why it exists, switch it off.
Where to look
Automations hide in the tools you already pay for. Spend an hour on four places:
- Your field service and booking apps. Look in the settings for anything called automations, workflows or integrations.
- Your email and calendar account. Look for the list of connected apps that have access to your account.
- Your card statement. Small monthly charges for automation or AI tools are often the first clue that something is still running.
- Your people. Ask each person what they have set up or connected, including tools they tried once and forgot.
Write down everything you find, even the ones you plan to remove. The list is the point.
What this changes
The hard part of automation is no longer building it. It is knowing what is running, who answers for it, and how to stop it on a bad day. A business that can answer those three questions can add new automations calmly. A business that cannot is adding to a pile.
When you hire someone to build an automation, ask them to fill in this row for you before it goes live. If you want help with that, here is what we build.
Your one step this week: block one hour, start the spreadsheet, and find one automation nobody can explain. Turn it off or give it an owner.
Sources
- Dataiku launches Agent Management to track and manage performance of AI agents built and running on leading platforms — Dataiku via Business Wire, September 24, 2026
- Discover Dataiku Agent Management: every agent on the record — Dataiku, September 24, 2026
- Global AI Confessions Report: CIO Edition, announcement — Dataiku, September 24, 2026
- Dataiku debuts cross-platform Agent Management, expands Cobuild building agent — SiliconANGLE, September 24, 2026
Kush AI Automation builds in accounts you own, and our monitoring alerts when a run fails and when it never starts, so an automation that quietly stops still reaches a person. Bring one process you want handled and book a free call.
