NORA examines evidence with a magnifying glass beside a colleague and an automation screen.

AI can already act. Who is ready to supervise?

7 days ago

5 min read

Market Signals | Edition of 25 October 2026

Get your favourite drink and join the conversation. ☕ Imagine an AI demonstration has just ended. The task moved forward, the result appeared and someone is already thinking about everything they could automate on Monday.

Then comes the question that makes everyone in the room look at the same person: “And who will keep an eye on this afterwards?”

That is the moment we are going to talk about. When the novelty meets the daily routine, someone needs to look after what happens between pressing the button and receiving the result.

The demonstration ends. The work continues.

NORA points to an empty chair before starting an automation.

In our imaginary scene, nobody had forgotten about oversight. It was simply in that well-populated project category: “we’ll sort it out later”.

The problem is that later arrives. The team needs to use the automation, understand what it did and know whom to turn to if something unexpected happens.

Delegating a task also calls for a conversation about responsibility. Who monitors it? Who can intervene? And who gets the call when the result looks odd? 🤔

NORA may admire the demonstration. But she is already looking for the name of the person who will answer the phone.

🧭 NORA Guidance | Before expanding use Agree who monitors the automation and who can intervene. Make this clear to the people who will work with it. “Someone will keep an eye on it” still needs a name.

From the promise to Monday’s work

NORA compares three perspectives: the offering, leadership and professional responsibility.

An assistant needs boundaries too

Where did this story come from? From the article introducing ServiceNow Arc by Paty Montesinos, identified as a ServiceNow Employee on the Community.

In the introduction to Arc, the proposition of an enterprise assistant comes with a discussion of governance: the rules and responsibilities surrounding its actions.

Think about what that means for someone assessing a new offering. Alongside curiosity about what it can do, it is worth asking: within which boundaries?

It is a useful conversation to have while everyone is still excited about the demonstration. Enthusiasm can join the meeting. It just does not need to sign the authorisation on its own.

And what if nobody can see clearly?

Where did this story come from? From ServiceNow’s announcement about its research with Phronesis Partners, published on 20/10/2026.

The announcement points to gaps in accountability, visibility and the ability to intervene in AI agents’ actions.

To put that into an everyday situation: imagine having to explain an automated action and discovering that each person knows only part of the story. One authorised it, another received the result, a third knows where to find the record.

Nobody needs to memorise everything. But being able to reconstruct what happened makes a difference in that conversation. “It must be somewhere” is a very popular and rather unhelpful location. 🙂

AI does the task. Who keeps an eye on it?

Where did this story come from? From Unit4’s ServiceNow Developer advert on the recruitment platform used by the company.

In the role, working with Now Assist and AI workflows also involves controls, testing, monitoring and human oversight.

Here, the conversation reaches the desk of the person who will develop and monitor the solutions. The role description already includes this care alongside automation.

AI can take on a task. The familiar “we’ll deal with anything else later” still does not qualify as an oversight plan.

Keep this image in mind: the presentation finishes, the meeting ends and someone opens their computer to handle the real work. That is the person with whom it is worth agreeing responsibilities from the start.

What happens after the button?

NORA and a colleague check authorisation, records, intervention and testing.

Let’s imagine an automation preparing an update for other people. The text looks good, the data is organised and everything is ready to go.

Before letting it act, the team agrees what it may do, what needs to go through a person and which record will allow the work to be checked. If the result looks odd, they also know whom to call and who can stop execution.

Notice how the conversation has changed: we are already talking about the conditions for using it responsibly. Testing and monitoring come with delivery, with attention to the people who will receive the result.

For people working in development or architecture, there is an opportunity to make this understandable to colleagues. Explaining boundaries, listening to users and agreeing how to handle deviations are also part of a good conversation about technology.

The button says “automatic”. The agreement between people deserves a few more words.

🧭 NORA Guidance | A conversation worth having Before expanding use, speak to the people who will operate the automation and receive its results. Include testing, monitoring and intervention in that assessment. Also look for accounts of use with documented results and failures.

What do we take away from this conversation?

NORA talks with two colleagues around a table with mugs and a checked list.

In this edition, oversight appeared in the presentation of an offering, in a security concern and in a role description. For our conversation, one thought remains: alongside autonomy, it is worth agreeing who looks after it.

Because after the demonstration comes Monday. There is a team, a task waiting and someone who needs to trust the result to get on with their own work.

Let’s hope that person receives useful automation — and a better agreement than “if anything happens, call someone”. ☕

Sources and editorial transparency

NORA checks documents and a source notebook with a magnifying glass.

Paty Montesinos — introduction to ServiceNow Arc. Article by a ServiceNow Employee; vendor-linked communication, without evidence of customer results. The current page has been updated; this edition preserves the connection recorded in the analysis of 25 October, without incorporating new specifications as facts from that period.
ServiceNow/Phronesis Partners — research on agent security, announcement of 20 October 2026. Research involving the vendor; it is not independent validation of the effectiveness of its offering.
Unit4 — ServiceNow Developer, requisition 744000147956820. Employer’s description, without a confirmed first publication date. The advert does not establish a new vacancy that week, a completed hire or adoption of Arc.

Edition based on the integrated analysis of 25 October 2026 and the Radar cycle of 24 October, with different cut-offs between the evening of 17 October and the evening of 24 October. Community and Jobs & Hiring had partial coverage; C-Level was complete within its targeted sample. Public sources were consulted again on 27 October 2026, without extending the original window.

The three perspectives have limited independence: Community and C-Level share a ServiceNow origin for part of the evidence. No conclusion is drawn about market growth, product purchases or the effectiveness of controls. Native behaviour, licensing, availability by release and operation in an instance were not verified. The team scenes are illustrative; the guidance represents NORA’s interpretation and recommendations.

Content consolidation and writing support: NORA — Market Signals.

Editorial review and final approval: André Costa Moreira.