Automation governance ยท 5 min read
Automations without an inventory: knowing what runs and who owns it
By FlowSharp
November 18, 2026
Scripts, integrations and automated flows keep multiplying, and when one breaks the customer is often the first to notice. The issue is inventory and ownership, not technology.
The problem is the missing list, not the technology
A script written to fix a problem on a Friday afternoon. An integration between the ERP and the CRM. An automated flow set up by a colleague who has since changed role. Each of these made sense when it was created. The trouble shows up later: there are dozens of them, maybe a hundred, and nobody has a list saying what they do, where they run and who answers for them.
The symptom is familiar: something breaks and the customer is the one who notices. A confirmation did not go out, a case is stuck, a piece of data never arrived. At that point the question is no longer technical (why did it fail?) but organisational: whose automation is this? Who should have noticed?
For IT teams and integrators this is nothing new: what gets built quickly gets documented slowly. The answer is not less automation. It is knowing what exists, what it has done and who is responsible when it stops working. It is the principle behind FlowSharp: measure first, then automate.
What Gartner says, and where the parallel ends
According to Gartner's press release of 28 April 2026 on managing the proliferation of AI agents ("AI agent sprawl"), presented at the Digital Workplace Summit in London, by 2028 the average global Fortune 500 company will have more than 150,000 agents in use, up from fewer than 15 in 2025. Gartner also reports that only 13% of organisations believe they have the right governance for AI agents. Among the steps it recommends is a central inventory of agents.
Two caveats matter. The 13% is a self-assessment: it tells us how many organisations feel prepared, not how many actually are. And the release is about AI agents, not about automations in general, so it should not be read as a figure on scripts, integrations or traditional flows.
We therefore use it as a parallel, not as proof. If a central inventory is the starting point for AI agents, the same question applies to the wider set of automations that nobody governs explicitly. The figure describes the problem; it does not show that any particular tool solves it, FlowSharp included.
An inventory starts where automations live
In FlowSharp, automations live inside Flows as tasks: Webhook, Integration, Event, Script, AI. They sit in phases that have owners and rules, and they are visible in the designer, in the flowchart and in the case detail. The list of what can fire is therefore not a separate document that goes stale; it is part of the process definition itself.
The phase owner is a first point of reference for accountability: whoever looks at the flowchart can see where an automation acts, in which phase, and with which rules around it.
There is a limit worth stating. This visibility covers what is inside the Flows. Scripts running elsewhere stay outside this list and have to be catalogued by other means; bringing them into the Flows is a process decision, not something that happens automatically.
To start from what already exists, Settings > Diagnostics lists the tasks that need fixing, for example Event tasks that publish restricted events. It is a practical work list, not an overall verdict on the state of your automations.
Who acted and what went wrong
Knowing an automation exists is not enough: when a case behaves unexpectedly, you need to reconstruct who did what. The case history shows who acted: a person, an automatic rule, a connected assistant, an integration key or the engine. The history exposed through the API and MCP carries a structured author, which helps anyone reading this data with other tools.
For events the distinction is finer. An event stays in progress while there are deliveries to complete or retry, and it separates successful deliveries, permanent failures and skipped sends. A delivery that runs out of attempts is closed as failed, with a readable reason. A failure leaves a trace a person can interpret, instead of a silence discovered after the fact.
On the AI side, the AI call log (Admin area) records provider, channel, model, case, tokens, latency and outcome. For whoever has to answer for how assistants are used, it is a starting point for questions about what was called, on which case and with what result.
Who picks it up when something stops
Visibility only helps if someone receives the signal. FlowSharp shows an "Automation stalled" alert when a case sits in an automatic or AI phase without moving. It also flags cases where the chain of automatic transitions does not stop by itself, which points to a probable loop in the rules, or where nested sub-processes exceed the limit.
An Integration task can also hand over to a person after a permanent error, after the planned attempts, or when the response is not among the expected outcomes; the case then shows the Fallback badge. We covered how to design that link in the October 13 article on webhooks, retries and human fallback. Here the point is different: for that handover to work, someone has to be responsible for picking it up.
When an intervention outside the normal path is needed, exceptional recovery of a case requires an Admin with a stated reason and is recorded in the case history. Even the exception leaves a trace.
An alert reports the situation but does not decide who handles it: that is an ownership assignment to make beforehand. To get started, three questions are enough, asked for each process area: what is running, who answers for it, and how would we know if it stopped. Even partial answers are already a first inventory.