ServiceNow, which sells the software behind the IT help desks, HR requests and service queues at most large organisations, has launched a system that looks for broken business processes and then rebuilds them using its own AI agents. The company calls it AI Workflow Factory and made it available globally on 6 October. The pitch is that improving a workflow stops being a project a company runs occasionally and becomes something running all the time.
The loop has three stages. Process mining decides which processes need changing and ties that to a business measure. An Autonomous Engineer and a Build Agent then construct the change and check its quality, and App Engine runs the result at scale.
Process mining is the least familiar of those and the most important. Every enterprise system writes a log of what happened and when, and mining software reads those logs to reconstruct how a process actually ran, which is routinely different from how somebody drew it.
AI Control Tower, where an organisation watches what its own AI is doing, sits around that loop, and Action Fabric extends the same supervision to agents built by other vendors. That second part matters, because almost no large company runs agents from a single supplier.
Accenture and Infosys are the launch partners, with Infosys bringing its Topaz and Cobalt offerings alongside.
"The next phase of AI demands connected execution, not fragmented solutions," said Bhaskar Babu of Accenture, while Amit Zavery, ServiceNow's president and chief product officer, said the Factory "brings the full cycle of reinvention onto the ServiceNow platform."
A $1B AI Line To Protect
The launch sits on top of a business that has already sold a great deal of AI, which explains both the confidence and the urgency. ServiceNow's annual contract value from AI products passed $1B in the second quarter, against a target of $1.5B by the end of this year. Annual contract value is simply the yearly worth of signed subscriptions, and the company said the number of customers running agentic AI in production had grown ninefold over nine months.
The wider figures are large. Subscription revenue reached $3.877B in the quarter, up 24.5%, with 658 customers now spending more than $5M a year each.
Zavery's own account of how customers got there is more revealing than the growth rate.
"A lot of customers started with a piecemeal approach," he said, "then ran into security and governance issues," and he noted that most of the AI growth comes from organisations already running ServiceNow rather than from new ones.
That shapes what the Factory is. It sells more AI into customers who have already bought the platform. It also answers the specific complaint those customers made after their first attempts went wrong.
It Does Not Lead On Process Mining
The first stage of the loop is the one ServiceNow is weakest at, by the assessment of the analyst firm it competes on. Gartner evaluated 13 process intelligence providers this year and named four as Leaders: ARIS, Celonis, Pegasystems and SAP Signavio. ServiceNow was assessed and did not appear among them.
That matters because mining is what decides which process gets rebuilt. A loop that starts from a weak reading of how work actually flows will build the wrong improvement very efficiently.
ServiceNow's answer is structural rather than technical. Celonis and SAP Signavio can tell a company what is broken, but the fix then has to be built somewhere else. ServiceNow owns the system the work runs in, so it can carry the finding straight through to the change.
That is a real argument, and it is a different argument from leading on the mining itself. Buyers choosing between the two are weighing better diagnosis against a shorter distance from diagnosis to repair.
Gartner Sees Most Of These Cancelled
The forecast hanging over every launch of this kind is unusually blunt. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, citing rising costs, unclear business value and weak risk controls. It also expects 40% of enterprises to demote or switch off autonomous agents after governance problems surface in production.
The same analysis puts a number on how much of the market is real. Of the thousands of vendors describing themselves as agentic, Gartner counts roughly 130 that genuinely are, with the rest relabelling chatbots and older automation scripts.
What makes the forecast awkward for ServiceNow is that the causes Gartner names are the ones the Factory claims to fix. Poor process mapping, no baseline measurement and governance treated as a simple on-or-off switch are precisely what process mining, business measures and a control tower are sold to address.
So the product is a credible response to a documented failure pattern, and it is also unproven against it. The companies that cancelled projects in 2026 did not do so because nobody offered them a governance layer.
The governance question underneath is the same one every autonomous deployment runs into. An agent that rebuilds a workflow is changing how a company processes real requests, which puts where an agent's authority ends at the centre of whether the loop can run without a person reviewing each change.
Extending supervision to agents from other vendors, which is what Action Fabric is for, only works if those agents report what they are doing. That is the same problem of opening systems to each other every enterprise integration effort has faced.
What Would Settle It
The evidence will arrive in two places, and neither is the launch. The first is the annual contract value line. ServiceNow has put $1.5B on the record for the end of 2026, which is a figure its own results will test within months rather than years.
The second is retention among the customers already running agents in production. A ninefold increase over nine months describes how many started; whether those same deployments are still running in a year is the number that separates this from the pattern Gartner describes.
Neither is visible yet. AI Workflow Factory is available globally from today, the Autonomous Engineer is in early access on request, and ServiceNow has published no pricing for either.
