From the InnoValley board session and the Technion workshop, September 2026

AI from the management floor

Written: ספטמבר 2026Updated: 15.10.2026עברית

What happened

AI did not enter organizations through a steering committee. It came in through the browser. Someone on the team opened an account, pasted a draft and got a good answer. By the next week the whole floor was doing it. Within months the tools sat at the center of daily work, before anyone in management had decided on them.

The numbers tell the same story. In a Microsoft and LinkedIn survey of 31,000 people in 31 countries, 75% of knowledge workers said they use AI at work, and 78% of those users bring their own tools. In the same survey, 60% of leaders said their company lacks a vision and plan to implement it [1].

The view from the organization is similar. In McKinsey's 2026 State of AI survey, nearly nine in ten respondents reported regular AI use in at least one business function. 44% said AI is scaling across their enterprise, and only 37% attributed any EBIT impact to it [2]. Use is broad. Value is not there yet.

Where the problems are

Over the past year I have sat with management teams, boards and senior executives. Four issues came up in almost every room.

Data. Employees are already pasting documents into AI tools, and management does not know what, where or why. A contract, a payroll sheet, a client deck. Not out of carelessness, but out of a wish to move fast. Still, information leaves the organization without anyone having set a boundary.

Accountability. Management has to bring AI into the organization, and it is management that will answer for it. When use starts at the bottom, accountability stays at the top without the means to exercise it: no list of what is running, no owner for each use, no written decision.

Value. A new AI tool comes out every week, and nobody has asked which process it solves. Licenses are bought, training sessions are held, and at the end of the quarter it is hard to point to one process that changed or one metric that moved.

Thinking. The fourth issue is quieter. When the tool drafts, summarizes and recommends, it is easy to stop checking. The first draft looks finished, and it moves on.

What the research says about cognitive offloading

The phenomenon has a name: cognitive offloading. Evan Risko and Sam Gilbert defined it in 2016 as the use of physical action to alter the information processing requirements of a task so as to reduce cognitive demand. A shopping list is offloading. So is a phone reminder. There is nothing wrong with it in itself. They also showed that the choice of what to offload rests on our judgment of our own abilities, and that this judgment is not always accurate [3].

What generative AI changes is what we offload. Not only memory, but judgment. And here the findings are starting to accumulate.

A study by Michael Gerlich, published in 2025 in the journal Societies, surveyed and interviewed 666 participants. It found a significant negative correlation between frequent AI tool use and critical thinking, with cognitive offloading mediating the relationship. Younger participants showed higher dependence on the tools and lower critical thinking scores [4].

Researchers at Microsoft Research and Carnegie Mellon University asked 319 knowledge workers about 936 instances of using generative AI at work. Those with higher confidence in the tool engaged in less critical thinking. Those with higher confidence in themselves engaged in more. The researchers also described a shift in the nature of the work: thinking moves from creating toward verifying, integrating the response and stewarding the task [5].

At the MIT Media Lab, 54 participants were split into three groups that wrote essays: with a language model, with a search engine, and with no tool. EEG measurements showed the weakest brain connectivity in the language model group. Participants in that group also reported the lowest sense of ownership over what they wrote, and struggled to quote their own essays [6].

These findings need careful reading. They are early studies. One shows correlation, not causation. One relies on self-reports. One is a preprint with a small sample that has not yet been peer reviewed. They do not establish that the tools weaken thinking. They do point in a consistent direction: when checking drops out of the loop, thinking weakens. And in an organization, a decision nobody checked is a decision nobody truly owns.

What management can do

The answer is not to block the tools, and it is not another prompt-writing course. It is getting the order right. This is not a task for IT alone. It is a management decision, like any other investment, and it starts at the leadership table.

Goals before tools. Start with what the organization wants to achieve this year, and work down to processes. Which process hurts, who owns it, and what it costs today. Only then ask which tool fits. Sometimes the answer is a process fix with no AI at all, and that is a good answer. The output: investment priorities.

Rules for data. Define what goes into a tool and what stays inside, which information is sensitive, who approves a vendor, and where a person reviews before an action goes out. Every use gets an owner: someone accountable for the tool, for the data that goes in and for the output that comes out. The simplest rule is also the most important: what comes back from the tool gets checked, not accepted. Numbers, names, claims with a source, and recommendations that lead to action.

Pilots with success measures. Set a baseline before you start, define in advance what success looks like, and decide at the end: scale, fix or stop. Prove value before expanding.

These three moves put thinking back where it belongs. Whoever sets the goal, draws the boundary and measures the result keeps thinking, even while the tool does part of the work.

Where LEAD AI 7MS came from

The method did not start as a model on paper. It grew out of work with management teams, out of workshops, and out of a wide reading of existing frameworks for risk, readiness and process. I kept seeing the same pattern: the AI conversation opens with "which tool", and "what for" arrives too late, if at all.

So I reversed the order. LEAD AI 7MS enters the organization through management, with its goals and its accountability, not through the tool. It moves from understanding the organization and identifying pain points, through mapping data and risks, to fitting the solution, piloting, measuring and embedding. At every step, management sees what is happening, what it is deciding and what it gets. An idea that is not ready is not deleted. It stays on the map, with the reason and with what it still needs.

At the end, management holds three things: investment priorities, rules for data use, and pilots with success measures. I presented the method in September 2026 to the InnoValley board and mentioned it at a workshop for managers at the Technion. It is also available under license to firms that map organizations.

At Maya Shnur Advisory, I work with management teams on bringing AI into the organization, from the top down. maya-shnur-advisory.com