It takes a lot of courage to bring researchers into the absolute heart of your enterprise for almost a year. But when Natural Intelligence CTO Lior Schachter received a request from Tel Aviv University’s AI Lab to monitor the process of integrating AI into the company, he saw it primarily as an opportunity.
"I had nothing to hide. It was a way to get advice from fresh eyes on how to adopt technology correctly," he says, explaining the decision.
The Tel Aviv-based intent marketing company opened its doors to Prof. Lior Zalmanson, head of the Department of Technology, Management and Information Systems and head of the AI Lab, which researches management in the AI era. The field research was led by Dr. Maayan Cohen, an organizational anthropologist and Oxford graduate. As part of the EU-supported research, Dr. Cohen conducted dozens of personal interviews with employees, observed executive meetings, and mapped out the dynamics of the enterprise.
"In the first phase, it was mostly the Wild West," says Schachter about the start of the implementation process, more than two years ago. "We tried to outline a structured strategy, but it was an incubation phase, and the initiatives were mostly bottom-up-an employee could grab me in the hallway and show me something cool they had done with AI. There were also quite a few failures, which is great because we learned from them."
What surprised you most about the insights that emerged from the research?
Schachter: "We saw that it is not an organizational tool, but a personal tool. Everyone has their own practices, a way they create prompts that they strongly believe in. They are reluctant to share their prompts or copy prompts from others, because those don’t fit their persona."
Prof. Zalmanson: "It is a tool that creates a situation in which each employee actually has their own team. The result is that a lot of isolated islands are created in the organization where people work by themselves. Sharing knowledge in organizations has always been a challenge: employees want the knowledge to be identified with them, both for credit and recognition, and also because there is a concern that if they share it, they will end up being expendable."
And when talking about knowledge, we no longer mean the traditional knowledge required in the world of work.
"If in the past manual workers were valued, and 15 years ago the tension was around sharing knowledge, the challenge today is in sharing personal prompts and getting them adopted by the entire organization. This simply introduces a new source of tension," says Prof. Zalmanson. "It's not just a matter of security and job survival; it's simply because AI tools, like ChatGPT, are identified as an extension of the self. In other words, 'I did it thinking about what suits me, I'm not sure I'm proud of it in the sense that I want it to be something others learn from.' In fact, working with AI tools is perceived as something highly intimate."
More often than not, in the workplace, we simply want to appear like we have it all together. We don't want others to know how much we don't know. The advantage with a chatbot is that I can be completely honest and tell it how much I don't know. But I don't want my colleagues, or my manager, to see that.
Prof. Zalmanson: "It gets more complicated when it's a tool that companies provide to employees, like at Natural Intelligence where they supply an enterprise version of the AI tool, rather than a personal tool that the employees bring in themselves. With AI, you create a kind of intimate advisor for yourself, one that requires a certain level of vulnerability in order to succeed. That makes sharing and formalizing that same tool complicated."
Schachter: "It's a multidisciplinary tool that crosses into different areas. As a manager, you raise professional questions or dilemmas about one of your employees or a colleague. Just as you don't want people to look into your head and see all the managerial and personal conflicts you have, you also don't want them to gain access to your chat history."
Mediation that reduces friction in the organization
With AI, one moment you can be asking it about a piece of code you wrote, and the next you are using it as a consultant on how to tell your manager that you need to leave early today. For years, we have been accustomed to treating work as if professionalism was the main thing and human relations were secondary; now, AI is making human relations a significant part of the day.
Prof. Zalmanson: "Until now, employees did not realize how much they needed AI. Think, for example, about cross-cultural communication: you get an email from an employee in another country and you want to check that you understood them correctly, or make sure you didn't insult them with your response. AI allows for a type of mediation that reduces friction in the organization. We have known the concept of personalization for years in tech, but this is a form of hyper-personalization. It goes so deep that it becomes part of the individual, and this has major managerial implications.
"First of all, it makes the company highly entrepreneurial. It allows an individual employee to do 'vibe coding,' connect to databases, and perform data analysis-you can do a lot of things as a one-stop shop. But then the question arises: What happens in organizations where every single person becomes capable of such things? How do you manage a group of entrepreneurs?"
And does this make you think differently about management now?
Schachter: "In the first phase, we witnessed truly wild entrepreneurship, and even though it was highly personal and sometimes created isolated islands of prompt engineering, it produced a real impact. At the beginning of 2025, I stood in front of the board and presented an analysis showing that we had improved productivity by 9% using GenAI. That number represents the percentage of employees who said they saved between five and 20 hours of work a week with the help of AI."
At this stage, Schachter recounts, the company moved from an incubation phase to a standardization phase.
"We shifted to a way of thinking that is more top-down. We added another strategic layer that comes from above, targeting specific areas of the organization and pushing them toward automation. For example, we have automated all of our ad copywriting on Google and Bing. We are one of the largest advertising companies in the world, spending over $500 million on ads annually, and we have reached a point where all ads are generated by GenAI, using a standard for prompts and context.
"Then we integrated this into the tools of the marketing staff, who could choose from the suggestions they wanted for their campaigns. We ran tests and saw that ads created by the AI received a score from Google that was 25% higher than ads created by humans. These were our first attempts to look at what everyone was doing in the secrecy of their own GPT, identify the best practices, and elevate them for organizational use."
The transition to one tool: "Veteran staff came to my office to complain"
Do you have an example of something that didn't work in your AI implementation process? Schachter: "Yes. Code generation. The tools were introduced in 2023 and we handed them out to see how they accelerated processes. But it was very difficult to measure the impact, perhaps because programmers weren't enthusiastic about the tool or because the model wasn't good enough. The big change happened in mid-2025. There was another leap in the models and I started hearing more and more programmers say, 'Wow, this is starting to do the job.' Then we made a brave decision: to stay with just one tool, Cursor, implement an enterprise version, cancel subscriptions to all other tools, and move all teams to work exclusively with it.
"A lot of veteran employees came to my office and said to me: 'What have you done to us? I was fast and good, I loved what I do. Why Cursor now?' But even the toughest and most excellent employees, after two or three months, understood the value of consolidation and collaboration within and between teams. What allowed us to adopt this was mainly the ability to measure-once everyone is working with one tool, we can create a dashboard that shows how many lines of our code are written by AI. In August 2025, it was about 20%, and by January of the following year, we were already at 53%." Today, the company notes that the figure has already reached 90%.
Schachter attributes the success mainly to insisting on order and uniformity.
"There is a word that is not well-liked in Israeli tech, and it is discipline. But you should not be afraid to say it, because in the end, when an enterprise wants to get the most out of this and ensure employees aren't left behind, you have to standardize and share."
So we are hearing a tension here between the desire to encourage bottom-up initiatives-where employees bring in all kinds of AI tools and show what they can do-and the drive for standardization. The first part fits startups perfectly, but is it suitable for large companies?
Prof. Zalmanson: "We have often been told, 'It's easy for you because you are researching a tech company.' But even in other fields like law firms, transportation, or Unilever, employees are already coming in today and showing how they can do more using the tool. The question is how management stays attentive to this and actually learns it. The problem is that in quite a few organizations today, we see that the leadership isn't learning enough about the field.
"One of the fantasies managers have is, 'We'll do an AI project in 2026, and next year we'll move on to another project.' But this is not a one-time, 'one-and-done' change; it won't go away. Resources need to be permanently allocated to this field, and AI Champions must be appointed within the organization. We have an AI leadership course at the university that teaches people the tools, but first and foremost, it teaches them how to diagnose-where there should be automation, where it is better to deploy an agent, and so on."
Schachter: "Ultimately, when I look back, this is a process of true organizational transformation-rethinking team structures, hierarchies, and role definitions. Such a change cannot work if there is no strong commitment from the managerial echelon, alongside the seeds that sprout from below. There have been quite a few moments over the last two years when certain areas of the organization almost gave up and said, 'The tool doesn't work well enough, the output isn't good enough.' But giving up is not an option."
Okay, so I understand how AI is implemented, but as a result, are managerial roles themselves actually changing?
Prof. Zalmanson: "Another role has been created for managers, and that is to manage employee trust. We are moving between two anxieties: on one hand, that there will be underutilization of the tools and we will miss out on amazing capabilities; and on the other, that there will be blind reliance. We see in the lab that people tend to agree more and more with the biases of AI, and here lies an important role: managing the competence of employees in this field and keeping them alert and committed. More and more tasks will be automated, and humans will shift into oversight and control roles.
"Which brings me to the second insight: we need to start treating our employees as managers themselves. Meaning, if my employee has to act as a supervisor, perform quality control, delegate authority for tasks, and execute prompt engineering-all of these are core managerial capabilities. Today, we need MBA studies more than ever before, because everyone needs to know how to manage. Responsibility filters down one level, and the managers at the top now have to manage the management processes of their own employees."
Published by Globes, Israel business news - en.globes.co.il - on May 19, 2026.
© Copyright of Globes Publisher Itonut (1983) Ltd., 2026.