A control tower is not a map of your operation

The enterprise software vendors have noticed that agents are not reaching production, and they are answering with governance: control towers, audit trails, guardrails. That helps. But a control tower tells you what your agents did. It cannot tell them where the work is. That map has to come from the operation itself.

What the vendors announced

On 6 October, ServiceNow launched AI Workflow Factory. Per its own press release, the loop connects process mining, its Autonomous Engineer and Build Agent, and App Engine to run the improved workflows, with AI Control Tower governing the workflows, decisions and agent actions. On 21 October, Oracle announced AI Agent Studio for Industry Applications, a no-code and low-code environment for creating, orchestrating and managing agents for industry-specific business processes.

Read the shape of those announcements. The market is moving from "here is an agent" to "here is a loop: find the work, build the automation, run it, govern it." That is a healthy correction. It is also a quiet admission that the agent was never the hard part.

Governance answers the wrong question first

Governance asks: was this action allowed, and can we audit it? Operations asks something earlier: which work should exist at all, who owns each hand-off, and what does done look like?

If you answer the first question without the second, you get a well-audited agent doing the wrong job. The log is perfect. The queue is still full.

This is the same trap as buying the tool before redesigning the process, one level up. Now the tool comes with a dashboard, so it feels like control.

Mining tells you what happened, not what should

Process mining is useful. It shows where work waits in the systems that record it. Two limits matter:

• It sees only what is logged. The hand-off that happens on a personal WhatsApp, the spreadsheet someone keeps "just in case", the decision made in a hallway: none of that is in the event log.
• It describes today's process. If today's process is a workaround, mining makes the workaround visible, then optimizes it.

Someone still has to decide what the operation should be. That is design work, and it is done with the people who run the work.

What it looks like when the map comes first

The OCESA reporting operation is a good example of the order that matters. Ticket sales, sponsorship results, ratings and media evidence lived in different places. The work was to connect the evidence to the question, and the result was −70% reported operating cost, a normalized comparison from the project case record. Nobody started with a control layer. They started with the question the reporting had to answer.

At Rodo, WhatsApp, social and web enquiries were brought into a service operation with a measurable resolution path. In the first measured window, 82% of live-created enquiries were resolved (cohort 31 August–24 September, from the Rodo operational report of 24 September 2026; figures describe pipeline records, not all conversations). The measurement was possible because the path to resolution was designed to be visible, not because a tower sat above it.

What to do this week

Before you evaluate any governance layer for agents, answer three questions on one page:

1. Which single workflow would you hand to an agent first, and who owns its outcome today?
2. Where does that work wait, including the places no system records?
3. What would "done" look like, in a number you could check next month?

If you cannot answer them, a control tower will give you a very good view of a very unclear operation. Draw the map first.