From Zero to 3,500 AI Agents in Two Months

Jun 1, 2026 | minute read

When people and machines become colleagues, leaders face new challenges: Who governs the agents, what should we prioritize, and how do we ensure that what we do actually creates value? 

Last winter, we said it was green lights and full speed ahead on everything related to personal productivity, and we went from zero to 3,500 agents in two months. That’s when we realized we needed a marketplace for overview and sharing of agents, says Hege Skryseth, Executive Vice President for Technology, Digitalization and Innovation at Equinor.

Greater personal efficiency and smarter ways of working with AI are valuable – but the real impact comes when AI becomes part of core business processes.

This places new demands on leaders. As AI agents become part of the organization, leaders are increasingly caught between speed and control, governance and freedom. To succeed, organizations need alignment across technology, finance and people.

Why Leadership Alignment Matters  

That is why Skryseth recently met Even Westerveld, Executive Vice President for People and Communication at DNB, and Ronny Jevard, Chief Financial Officer of Sopra Steria, in Dagens Næringsliv’s studio. There, they shared experiences about the importance of the interaction between precisely these three roles. 

Even Westerveld says that one of the key steps to bringing the organization along is to raise AI literacy among the organization’s leaders. 

We put all our leaders back in the classroom to learn “vibe coding” – as a symbolic move to show how important it is that leaders understand the power of artificial intelligence, he says, and continues: 
It is unrealistic to expect all leaders to be AI experts, but they need enough understanding of the technology, our tools and our processes to know where we should prioritize our efforts.

Watch the DN Studio broadcast here

Leaders Balancing Speed and Control  

DNB has invested significant effort in raising employees’ AI skills to a minimum level. Now the company is working more purposefully to create value. 

We have employees who have 300 agents reporting directly to them, and that challenges how our organization is set up. If AI is to create real impact, we need to redesign our processes and rethink how simple we can make things, says Westerveld.

Cross-functional teams at DNB are exploring and testing what artificial intelligence can achieve. They have more than 2,000 technologists to draw on, and much of the work is about simplifying everyday life for customers. For example, the bank is testing new AI features in the Sbanken app that give people a better overview of their spending. But AI is evolving quickly, and leaders need to keep up to stay ahead. 

Every other week I alternate between feeling that we are ahead of the curve – and worrying that we are falling behind. Just in the last few months, we’re having completely different discussions about what is possible, and that challenges us. We can’t sit at the top and make all the decisions; it’s too slow. We need to build frameworks the organization can run within, says Westerveld.

A Pace That Challenges Leadership Teams 

Sopra Steria CFO Ronny Jevard sees that many organizations are facing exactly the same issues, and that leaders today have to be far more flexible than before. 

We put a governance model in place internally five months ago, and now we are already revising it. That illustrates just how fast things are moving right now. Many organizations succeed in testing and experimenting, but struggle with what is much harder: scaling the solutions and creating real impact in operations, he says.

He points out that the technology is evolving so quickly that many leadership teams become uncertain about when to invest – and how heavily to commit. 

Many hesitate to take the big step because they know that in three months there may be new technology. But it is also a risk to wait too long. The organizations succeeding now are the ones that manage to build governance and prioritization while still keeping up the pace, says Jevard.

He sees this becoming particularly challenging in leadership teams where leaders move at different speeds and have different levels of risk tolerance.

When some people are a bit cautious while others want more speed, it is easy to end up with compromises that do not really lead anywhere. At some point, someone has to make the tough decisions, while also being aware that AI investments also mean cuts or deprioritizations somewhere else, he adds.

Impressive Technology is Not Enough

Greater efficiency in itself is not proof that the decisions you are making are the right ones. 

It is not difficult to find good AI ideas. The challenge is to prioritize which initiatives should be scaled – and to be sure they deliver measurable effects. We see many impressive technology projects, but not all of them create business value, says Jevard.

He emphasizes that increased capacity does not automatically mean the organization has created value. 

AI can create an enormous increase in capacity, but capacity is not synonymous with value. That is where many organizations face their biggest challenge. Organizations must reinvest the time they save in something valuable, or they must actually realize the benefit in cost savings. That is precisely where many encounter the toughest priorities – when they actually have to decide what the extra capacity should be used for, explains Jevard.

Do You Have Control of the Agent? 

The arrival of AI agents reinforces the need for control and governance – and people must remain in the driver’s seat. Even if something is technologically possible, you cannot unleash the technology without understanding the consequences. Strong governance and standards for which tools are allowed, who approves them, and how risk is managed are key. 

Skryseth says the agents have many positive qualities but also challenging aspects: 

They can start behaving in unexpected ways and need to be monitored constantly.

Equinor uses a traffic-light model to govern this. On green are agents for personal use. On yellow are simple agents for the productive environment, which require different licenses and better control, and on red: advanced agents embedded in core processes. 

We need absolute control on red. They may be agents monitoring production or performing tasks on our platforms, and this is where the leadership perspective comes through most strongly: you need someone who dares and are willing to challenge and redesign these processes, says Hege Skryseth.

Since 2020, Equinor has realized NOK 3.3 billion in value from AI, and agents are already integrated into critical operations. They raise entirely new questions. 

How do you govern the agents? More and more standardized orchestration platforms are emerging, but do you really want to make your data and work processes available in an open cloud? It is easy to get excited by the possibilities, but you need control, Skryseth points out.

Who is Responsible When Agents Act? 

Sopra Steria sees that many organizations underestimate what happens when AI agents move from being tools for employees to actually performing tasks on behalf of the business at scale. 

Building a single agent is relatively simple. The major challenge begins when hundreds or thousands of agents interact, and especially when they operate on behalf of the business. When agents start making decisions or acting in your systems, they can create a lot of value – but also significant risk, says Ronny Jevard.

He believes this raises entirely new questions for leadership teams: 

Who is responsible if an agent acts incorrectly, violates guidelines, or makes decisions it should not have made? Is it the employee who built the agent, the people who set up the systems – or management? Organizations need clear governance around what agents can and cannot do, and, importantly, they must know: who is responsible when something goes wrong, says Jevard.

AI Powering Both Cyberattacks and Cybersecurity

Artificial intelligence has proven to be an accelerator for cybercrime, fraud, and digital attacks. That places new demands on organizations’ security, expertise, and countermeasures. 

Around 80 percent of all cyber alerts against Equinor are stopped through automation and integrated AI, and DNB stops 9 out of 10 kroner in fraud attempts using AI-based security systems. 

DNB is experiencing increasingly sophisticated fraud attempts driven by advances in generative AI. Last year, DNB’s CEO and CFO were exposed to so-called “CEO fraud,” where faces and voices were cloned in a highly convincing way during a fake digital meeting. Now leadership discussions at DNB are shaped in part by Mythos, the AI model that can detect vulnerabilities no one has managed to find. 

The only way to meet these attacks is with the same technology. But security is still very much about people. Everyone must be trained to understand risk, and we have mandatory training for both employees and subcontractors, says Hege Skryseth. 
We Must Talk Honestly About AI

DNB is experiencing increasingly sophisticated fraud attempts, accelerating alongside the development of generative AI. Last year, DNB’s Group Executive Vice President and CFO were exposed to so-called “CEO fraud,” where faces and voices were cloned in a highly convincing way during a fake digital meeting. Now leadership discussions at DNB are shaped by Mythos, the AI model that can detect vulnerabilities no one has managed to find. 

This means all organizations will have to focus more on digital security going forward. And while many fear that AI will replace jobs, I believe it will also create many new ones. Tasks will change hands and roles will disappear, but at the same time there will be a major need for new skills, says Even Westerveld.

He believes leaders must be aware that not everyone dares to express their skepticism about AI development. 

Those who are skeptical may not be the loudest voices. Leaders need to be good at talking honestly about AI, both about the value – and about the dilemmas as well, says Westerveld.

People Gain Superpowers 

Equinor, DNB and Sopra Steria all see employees with great enthusiasm for AI tools – and not least for the range of possibilities the technology represents. 

And the people embracing AI most actively are definitely not only young “digital natives.” 

The people using AI most actively with us are those with deep domain expertise. Often the seniors who know our processes and tools well, and who are able to connect AI in ways that give themselves superpowers, says Even Westerveld at DNB.

Hege Skryseth sees the same spark at Equinor. Employees delegate boring tasks to machines, while one skilled employee can solve an enormous project in just a few hours. 

An offshore colleague got access to a license on Friday. He woke up in the middle of the night on Saturday, sat down, and programmed a model giving a full overview of all the turbines in one of our wind farms – and that says something about the power. It took him four hours to create something that would previously have taken us several years. And he is not even a developer, but he has a great passion for the field, she says.

The Human Advantage

Ronny Jevard at Sopra Steria also believes that experience and AI are an unbeatable combination. 

Contextual understanding, ownership, and the ability to see things in multiple dimensions still come from people. I believe that will become even more important going forward. AI will not be able to help us with this – at least not yet, he says.

It is clear that AI can provide a tremendous advantage, and the three leaders predict that we may see a gap emerge between organizations that manage to change and those left behind. The same applies between the private sector and the public sector, where change often takes longer. 

Ronny Jevard believes a key to making this happen is understanding each other’s disciplines within the leadership team – across technology, finance and people. 

And then we have to dare to challenge both ourselves and each other and be willing to have the difficult discussions. Only then are we able to make the sound, cross-functional decisions required to succeed with AI, says Jevard, before ending with a comment from a leadership meeting that sums up where many people stand: 
It is extremely difficult to achieve, but it would be crazy not to try.