AI has entered the meeting room. Now who is accountable for the costs and decisions?
NORA Radar — C-Level | 18–24 October 2026
Enjoy the read! Get your coffee or favourite drink ready, because today’s conversation begins in a meeting room. ☕
Picture the scene.
A team presents a new artificial intelligence solution. It understands requests, organises information, consults systems and carries out tasks without needing instructions at every step.
The demonstration is impressive. ✨
Until someone asks:
“How much will this cost when thousands of people start using it?”
Someone else wants to know who is accountable if the agent takes an inappropriate action.
And a third person asks an even simpler question: “How will we know whether this has actually improved our work?”
The technology may work perfectly during the presentation. That does not mean the company is ready to use it.
This scene is illustrative, but the questions are real.
In the C-Level Radar of 24 October, publications about artificial intelligence, security, costs and human work offer five perspectives on what is at stake.
Two of them directly involve announcements and research released by ServiceNow. The others address broader business priorities, without establishing purchasing decisions involving the platform.
Let’s explore why these stories deserve attention.
Investing more in AI does not guarantee better results
Where did this story come from?
Source: Gartner — More AI Investment Is Not Creating Enough AI ROI
Date: 19/10/2026. Nature: an analysis of business research by Gartner, with commentary from analyst Alicia Mullery. This is not an announcement of ServiceNow adoption.
Gartner analysed responses from more than 2,000 technology leaders and found a striking result: only 13% of the organisations surveyed reported achieving significant or transformative value from AI.
That does not mean the other 87% gained no benefit at all. It means substantial business results remain difficult to achieve.
From demonstration to results
Imagine a company that automates the classification of requests.
Previously, someone had to read each request and route it to the team responsible. Now artificial intelligence does this in seconds.
It looks like an obvious improvement.
But it is worth investigating: are requests being resolved more quickly? Have errors decreased? Are employees receiving better service?
And what does it cost to keep the solution running?
This is where ROI — Return on Investment comes in. In simple terms, it is the relationship between the financial benefit achieved and the amount invested.
Gartner observed that organisations achieving better results use more disciplined mechanisms to track costs, work with business teams and stop initiatives that do not justify continued funding.
The lesson is not to abandon artificial intelligence.
It is to recognise that automating a task and improving the business are different things.
🧭 NORA Guidance | Value before scale Define the problem AI should solve, how to measure the outcome and which costs will continue during operation. The question is not only “does it work?”, but “does the benefit justify maintaining and expanding it?”.
And when technology starts taking action without constant human oversight, another concern arises.
When AI acts on its own, who remains accountable?
Where did this story come from?
Source: ServiceNow — Ungoverned Agentic AI Exposure Is Compounding Inside Enterprises
Date: 20/10/2026. Nature: a corporate announcement about research conducted by ServiceNow with Phronesis Partners. The source explicitly involves ServiceNow.
The research surveyed more than 3,700 security and technology leaders across 20 countries.
According to the findings released, 90% of the organisations surveyed reported at least one security incident involving AI agents in the previous year.
However, only 13% had programmes considered defensible against the study’s five criteria, including accountability, visibility and auditability.
This figure of 13% should not be confused with Gartner’s. These are separate studies, with different participants and criteria.
The issue goes beyond what AI knows
Imagine an AI agent authorised to consult ticket information and update certain records.
It can perform useful tasks.
But what happens if it tries to access confidential information? Or modify a record that should remain protected?
Who set the boundaries? 🤔
Who can monitor its decisions?
And, above all, who can stop its actions?
The research released by ServiceNow helps explain the importance of AI agent governance: defining permissions, responsibilities, monitoring and responses when behaviour falls outside expectations.
The announcement also presents ServiceNow initiatives involving autonomous security and agent control.
But there is an essential distinction: a capability announced by a vendor does not establish that it is installed, licensed or operational in every organisation.
🧭 NORA Guidance | Autonomy needs boundaries Before entrusting important tasks to AI agents, establish which information they may access, which actions they may perform and how the organisation will monitor and stop inappropriate behaviour. Autonomy without accountability does not eliminate risks. It may simply make them less visible.
In the same week, ServiceNow itself presented an initiative to strengthen control over autonomous agents.
ServiceNow and NVIDIA: more autonomous agents, with rules
Where did this story come from?
Source: ServiceNow Newsroom — ServiceNow delivers personal enterprise agent and policy enforcement with NVIDIA
Date: 21/10/2026. Nature: an official corporate announcement from ServiceNow about its collaboration with NVIDIA.
The announcement introduced the general availability of Arc, described by ServiceNow as a personal AI agent capable of carrying out enterprise work over extended periods.
The proposition goes beyond answering questions.
An agent of this kind can receive an objective, plan activities and execute steps across tools and systems, within the applicable capabilities and restrictions.
But the most interesting point for businesses is not autonomy alone.
It is control over what that autonomy allows the agent to do.
A rule for each type of action
Think of an employee asking AI to consult documents, produce a report and update some records.
Consulting the information may be permitted.
Preparing the report may be permitted too.
However, changing critical production data may require a different authorisation.
This distinction helps explain ServiceNow’s announced approach: establishing enterprise policies in Super Intelligence Control Tower and enforcing restrictions while the agent runs, using mechanisms such as NVIDIA OpenShell.
The announcement also described the integration of AI Passport with NVIDIA technologies to validate permissions and restrict actions.
There is an important point about availability here:
On 21 October 2026, ServiceNow announced Arc as generally available in the context of EmployeeWorks, while broader AI Passport enforcement through the NVIDIA OpenShell and BlueField layers was planned for a future release.
It would therefore be incorrect to treat the entire announced solution as already available in any environment.
And operational outcomes described as goals should not be confused with proven results from customer implementations.
🧭 NORA Guidance | Control proportionate to risk Distinguish between tasks involving information retrieval, creation, modification and critical actions. Confirm which controls are actually available and which depend on future releases. Trustworthy AI needs to know its boundaries — and the organisation needs to be able to enforce them.
Security and autonomy are part of the same decision. But the bill does not end there. 🧾
The cost of changes the company did not choose
Where did this story come from?
Source: Origina — Large enterprises report spending more than a quarter of their software budgets on changes they would not have chosen
Date: 20/10/2026. Nature: research commissioned by Origina and released in a report with Thinkers50. It does not demonstrate behaviour specific to ServiceNow.
Origina presented a study involving 2,018 IT decision-makers in the United States and the United Kingdom.
According to the research, the organisations surveyed reported allocating, on average, 28% of their software budgets to upgrades, migrations and licensing changes they would not have chosen of their own accord.
The company uses the term Software Tax to describe these recurring expenses.
It is not a government tax, but a way of representing costs associated with certain vendor decisions.
When an upgrade changes every priority
Imagine a company whose systems are working properly.
An upgrade required by a vendor comes along.
The team needs to adapt integrations, allocate staff, test changes and put other activities on hold.
The upgrade may be necessary and bring benefits. The problem lies in being unable to choose when to carry it out or understand its total cost in advance.
Now imagine an artificial intelligence initiative that depends on those same systems.
The investment is no longer just about the product’s price. It also includes contracts, integrations, maintenance, staff availability and future changes.
This is what executives call TCO — Total Cost of Ownership: everything needed to acquire, operate and maintain a technology throughout its use.
An important caveat: the research was commissioned by a company offering independent software maintenance. This should be considered when interpreting the findings.
It would also be incorrect to claim that the percentages represent ServiceNow customers.
The broader business signal is that predictability, vendor dependency and freedom of choice are important aspects of investment decisions.
🧭 NORA Guidance | Understand the full cost Consider the initial price, operation, maintenance, integrations, contracts and the effort required to change direction. A sound technology decision needs to keep making sense after the sales presentation.
And there is another dimension that spreadsheets do not always capture adequately: the effect on people.
AI can improve a task and worsen the outcome
Where did this story come from?
Source: arXiv — When Better Opportunities Backfire: Service Worker Responses to Parallel AI Deployment in Customer Acquisition
Date: 19/10/2026. Authors: Xinyu Fu, Shulong Luo, Narayan Ramasubbu, Peng Shen, Xincheng Wang and Yucheng Wang. Nature: an academic study released as a preprint, with peer review not yet confirmed by the source.
The researchers analysed more than half a million sales contact records at an online health insurance brokerage in China.
The study investigated an interesting situation: AI voice agents began handling some sales contacts, while human workers received opportunities considered more promising.
The expectation was that the new division of work would improve people’s performance.
But the data analysed showed something different.
Human behaviour changes too
During the first three contacts of the day, the probability of workers receiving more promising opportunities increased by around 23 percentage points.
At the same time, customer acquisition by human workers fell by approximately 5 percentage points during those contacts.
In the researchers’ interpretation, the change in how opportunities were distributed also altered workers’ willingness to persist with certain contacts.
In other words, AI changed the organisation of work and, with it, people’s behaviour.
The study does not show that AI always reduces productivity. Its findings concern a specific operation and context.
Even so, it raises an important question.
When we implement automation, are we assessing only the system’s performance, or also what happens to the people working alongside it?
A team may gain speed while also facing new difficulties with coordination, motivation or quality.
That is why measuring only how many tasks have been automated may offer an incomplete picture.
🧭 NORA Guidance | People are part of the outcome too Track the quality of what is delivered, changes in team behaviour and the experience of the people doing the work. The best automation is not necessarily the one that performs the most tasks, but the one that contributes to better results.
What do we take away from this conversation?
Let’s return to our imaginary meeting.
The opening questions now seem to make even more sense.
Gartner showed the importance of turning investment into measurable value.
The research released by ServiceNow highlighted security risks and governance gaps involving AI agents.
The announcement with NVIDIA presented an approach to strengthening control over those agents’ actions.
Origina brought costs and vendor dependencies into the discussion.
And the academic study reminded us that technological changes also transform human behaviour.
These publications do not form a single study and do not establish a combined statistical trend. Each has its own sample, nature and limitations.
However, they help frame an important business conversation.
It is not enough to ask what AI can do. We need to understand what it should do, how much it costs, how to control it and how people will be affected.
For those working with ServiceNow, architecture, operations, security or management, this broadens our responsibility.
Knowing the features remains important. But understanding the problem to be solved and assessing the consequences of our decisions is equally necessary.
Ultimately, technology should not exist merely to perform more activities.
It should help people work better, overcome difficulties and achieve results that make a real difference. 💛
And that may be the best question to take into the next meeting.
Until the next read! ☕
N.O.R.A. — Clarity, Security and Responsibility.
Sources and editorial transparency
This edition uses the publications below, released between 19 and 21 October 2026, as an educational selection for the week ending 24 October 2026.
Gartner — More AI Investment Is Not Creating Enough AI ROI — 19/10/2026.
ServiceNow — Ungoverned Agentic AI Exposure — 20/10/2026.
ServiceNow and NVIDIA — Personal Enterprise Agent and Policy Enforcement — 21/10/2026.
Origina — Software Tax Research — 20/10/2026.
Fu and colleagues — When Better Opportunities Backfire — 19/10/2026.
Transparency: the studies use different samples. The AI agent research involved ServiceNow; the Origina study was commissioned by the company itself; the academic paper is a preprint. The figures should neither be added together nor generalised to the entire market.
The professional scenarios used in this publication are illustrative. Recommendations labelled NORA Guidance represent editorial interpretation, rather than statements from the sources.
Content consolidation and writing support: NORA — Market Signals.
Editorial review and final approval: André Costa Moreira.