Hello, everyone. You are watching ToTheMoon. Artificial intelligence has gained access to your work. Who will be responsible for the decisions it makes? Hello again. This is ToTheMoon. We have a new author joining the channel and a new segment devoted to technology and artificial-intelligence news. For now, it will probably appear once every two weeks. Watch today’s episode and share your comments—not only about the author, but, more importantly, about the subject itself. We want to create even more useful content for people who want to go deeper into how AI can be used in their lives, and who want not merely to follow the trend but to understand what is really happening.
Every day, every week, and every month, there is more and more AI news. In reality, it stopped being only technology news a long time ago. The more I think about it, the more I understand that these stories are first and foremost about people, and about very different areas of their lives. Today I want to talk about AI and the person inside a company: how people’s work is assessed, how their effectiveness is measured, how decisions are made about their roles, their development, or whether they should be laid off.
And, more broadly, can a person understand why a particular decision was made, why a particular conclusion was reached, how to work with that conclusion, and whether it can be challenged at all? Several stories from the past couple of weeks caught my attention, particularly stories involving Meta and Google. Not because these companies are somehow unique. Today, much of what is happening can look as though it concerns only huge global
corporations. In fact, the opposite is true: news from global corporations often makes visible a problem that almost any business can face, or is already facing. In Meta’s case, for example, employees claim that digital performance metrics and AI tools influenced layoff decisions. To me, this is a much broader question than the way it is currently framed in the news cycle. The larger question is what happens when a person believes they are making a decision independently, while a system has already assembled the available options, the ranking, and the list before that person becomes involved.
On July 14, news broke of a lawsuit filed by twenty-six Meta employees in federal court in California. The employees allege that, while preparing mass layoffs, Meta used internal systems and digital metrics that may have disadvantaged people with medical limitations, as well as employees who were on medical, family, or parental leave. The lawsuit refers to productivity metrics, the use of AI tools, digital-activity data, and various algorithmically supported rankings. The logic behind the complaint is easy to understand.
If a system measures the number of actions an employee performs during working hours, then someone who was absent for a period of time—even for a lawful reason—will naturally appear lower in a performance or productivity ranking, with figures that look less favorable. Formally, that may appear neutral. In substance, however, the system is counting activity alone and may fail to account for legitimate reasons why a person was away from work. We are not drawing a conclusion about the case here, because Meta currently rejects the plaintiffs’ claim that layoff decisions were effectively made by artificial intelligence.
The company says that all final decisions were made by people, not by AI. And this is where the really interesting question begins: where exactly is the boundary?
In a real company, we are unlikely to reach a point where AI literally presses a “fire” button or a “hire this person” button. Yet the decision that a person ultimately carries out—to dismiss or hire someone—may already have been shaped by a sequence of actions, algorithms, analytical memos, and management decisions built on analysis produced by various AI tools. It is extremely difficult to identify the point at which AI either does or does not influence a human decision. If a person signs the final decision, does that automatically mean every preceding stage was correct and properly checked?
Again, the main issue here is not who is legally right in this particular Meta lawsuit. We will eventually learn what the court decides. Nor is the question whether AI should be prohibited from taking part in personnel decisions. The point is that responsibility often begins much earlier than the final decision itself. That early influence is easy to miss. Any algorithmic assessment requires exceptionally strong management discipline and careful attention long before the result appears.
What exactly should be measured? What counts as quality of work—quantity or quality? Who checks whether the system creates an advantage for some people or causes systematic harm to others? While reading the Meta story, I remembered Timur Bekmambetov’s film known in Russian as “Kaznit nelzya pomilovat.” It begins with a detective, played by Chris Pratt, appearing before a judge while accused of murdering his wife. He has either ninety minutes or ninety hours—I honestly no longer remember—to prove his innocence.
What matters is that the judge is not actually a person. Her name is Madison, but she embodies artificial intelligence: an ostensibly ideal machine that judges a human being and must pass sentence. The film came to mind not because it made an extraordinary impression on me or because I am recommending it. What I remember very clearly is the physical chill I felt in its opening scenes. I thought: how can such an important decision in a person’s life be made by some supposedly ideal machine?
Even though it is a mainstream film, I felt an almost physical resistance, a sense that there was a fundamental error built into the premise. It felt wrong—and not because people are necessarily kinder or fairer. I think human beings can be many times more biased than AI and can certainly make far more brutal decisions. The difference is that a person can doubt their own conclusion. A person can notice an exception. A person assumes moral responsibility for the decision they make.
They may be able to analyze far fewer signals than artificial intelligence, but a machine can see only what can be translated into data, rules, and permitted options. That structure may leave no room and no legitimate basis for an exception. It is a very fine line. AI can see facts and patterns, but a decision about a person’s fate should not become an automatic extension of that person’s digital profile. Continuing with the news of the past week, I did not initially connect the Meta lawsuit
filed by twenty-six employees with a separate story involving Google and Android. Then I started thinking that both stories are probably built around the same underlying question. If access to information inside a company can influence a decision about a particular person, then access to data more broadly can do the same thing. Access to particular information determines how relevant and genuinely useful a model or AI assistant can become in its interaction with a person. These are different situations with different risks, but the principle is the same: with access comes influence over our lives.
That is what prompted this line of thought. On July 16, the European Commission adopted two sets of binding measures for Google under the European Union’s Digital Markets Act. The first concerns Android. Google must give competing AI services deeper access to eleven operating-system functions. In practical terms, a third-party assistant should be able to activate by voice, use context, interact with apps, and—with the user’s consent—perform actions such as booking a calendar meeting, ordering food, or drafting and sending an email.
The second part of the decision concerns search. Eligible competitors, including AI chatbots with search capabilities, are to receive access to certain anonymized Google search data. In other words, selected queries, clicks, and views will become available to other systems and help them develop their own search products. Google argues that the decision creates serious privacy and security risks for users. The European Commission, by contrast, says the measures contain the necessary security safeguards and will give users more choice.
In essence, this is an open, antitrust-oriented intervention. The changes will be introduced in stages beginning in January 2027 and continuing in July 2027. What matters to me in this story is that it exposes a different kind of competition. We are not talking about which model is smarter. We are talking about how important access to the operating system, applications, and the user’s context is if an assistant is going to interact with that user meaningfully. It is a struggle for access to information itself.
You can have an exceptionally powerful assistant, but without context that assistant is like a door without a key. OpenAI introduced ChatGPT Work on July 9. I am a fan of OpenAI in many respects, and at the same time I work extensively in
Codex, so I was very interested in taking a closer look at this new environment. I decided to discuss it in this episode because it strongly connects with what we have already been talking about: the point at which AI becomes naturally embedded in our lives and begins influencing them from many directions at once. AI can participate in a decision—or, more precisely, lead a person toward that decision. That raises the question of its role as an intermediary between a person and that person’s digital profile.
My experiment with ChatGPT Work is a useful example. I did not want to test it with an artificial task, because I already use both ChatGPT and Codex extensively for different kinds of work. I chose a real case that I understand well and whose result I could evaluate. I created a separate folder containing professional materials from several years. I admit that I selected the materials somewhat randomly, which made the result harder to assess, especially because I could not include confidential information or personal data.
I asked Work not merely to read the documents or show me a few trends. I asked it to reconstruct the development of my work over several years: how my priorities changed, which topics gradually became central, which questions I kept returning to, and what recurring patterns appeared across my professional activity. What overall direction emerged from these materials? I gave the same task to ChatGPT Work and Codex. My Codex environment was already configured. My first reaction was not a wow moment at all; it was mild disappointment.
I could not see the difference created by this new business environment, as OpenAI presents it. It felt as though nothing had changed. The result from ChatGPT Work was essentially the same as what Codex had already produced for an equivalent task in an equivalent environment. Then I understood why. I had been using Codex much more broadly than a typical scenario such as checking an error or organizing a folder. I had already connected large folders to it for nontechnical work.
It handled dozens of documents, and I used it to develop hypotheses, identify trends, and build presentations. So when I asked it to study a folder, analyze it, and produce something from it, Codex gave me the same kind of result it had given before because I was already using it as a research agent. I then made the experiment more demanding. I asked ChatGPT Work not to stay inside the folder, but to go beyond it and compare my conclusions with external sources. I wanted it to assess whether the strategy, path, and professional direction I had chosen aligned with the direction in which the broader market was moving, and whether the way my role as a talent-function leader had evolved corresponded to the current management agenda.
I was asking specifically about an international technology business. At that point the difference became genuinely visible, because Codex did not work with external sources. Yet even that is not the most important distinction. Saying only that Codex lacked external sources would be an oversimplification. Before Work, I personally connected several environments: I attached Codex to a folder, carried out separate research in ChatGPT, transferred the findings myself, and built the presentation separately.
I effectively acted as an assistant between AI tools, moving context from one to another. It was not exceptionally difficult, but it still required manual work and constant attention to what had to be remembered and transferred. Work, by contrast, keeps this entire cycle inside a single task. It uses the local context, supplements it with external information, performs comparative analysis, creates all the necessary dashboards and presentations, and, in effect, can keep the result up to date if your next task is configured accordingly.
That was probably the main outcome of the test: understanding how these workflows are connected, and where the differences and common ground lie between the two business environments OpenAI is promoting today. That is my first conclusion. I still cannot say how effectively I will be able to use it or how much it will help me in my work over time. But, honestly, I began by looking for a new feature and ended up realizing that this is more like a packaged form of a new way of working with AI.
That may prove interesting, but I want to test it over a longer period. Every time I discover a new wrapper or a new way of packaging a work process, another area opens up in which AI enters our daily lives—especially our professional lives. Current tools are fundamentally changing how we think, how we work, how we formulate tasks, and how we present them. At that point, this is no longer a question about OpenAI, ChatGPT Work, or Codex.
It is not a question of Codex, Claude Code, or Work as individual products. It is, of course, a question about labor markets. A particularly interesting development appeared in this context. On July 13, a short statement titled “We Must Act Now” was published. It was signed by more than two hundred economists and AI researchers, including, as I recall, about sixteen Nobel laureates. The statement contained several core points. Above all, it argued that AI could become radically more powerful over the next ten years—although, in my own view, we may see extraordinary development even within the next year.
This could produce such a profound economic transformation and such sweeping changes in labor markets that we need to begin taking specific action today. We need to prepare for those changes now. The statement does not dramatize unemployment, nor does it focus on the drama of job losses. Its emphasis, at least as I understood it, is different: it is about the effect of these changes on people’s lives, especially their professional lives. If people are to live through this transition safely and effectively, preparation has to begin today.
The current regulatory agenda is complicated. Some laws may constrain certain directions of development, but when it comes to the relationship between AI and people, AI is clearly moving much faster than today’s regulatory
institutions, rules, and potential legislation. Another revealing development highlights this influence, the scale of change, and the need to mitigate risks before they fully emerge. More than 4,500 Google employees signed a petition that was delivered to management on July 16. At least at the time this podcast was recorded, the company had not issued an official response. What is notable here? According to the president of Google’s union, this was the largest volume of employee feedback the company had ever received specifically on job security.
Employees are asking for guarantees of minimum severance, proposing voluntary departures with compensation before compulsory layoffs in different product divisions, and requesting that severance be available in forms such as continued paid leave, among other measures. In effect, they want guarantees in advance because they understand that the company’s organizational design and structure may change. The petition appeared at a moment when Google is increasing its investment in AI and continuing to reorganize teams very actively.
People are asking a very simple question: what rules will protect employees during this transformation? In that sense, we end where we began. These demands are not resistance to technology. The employees are not saying, “Stop AI and put everything back the way it was.” That matters greatly to me as someone who has worked with talent and people processes for many years. What people are really saying is this: if work is changing, let us agree on the rules in advance. If we can see what may happen in a year, a year and a half, or five years, let us think about it today.
Let us prepare now, because once those changes arrive, it may be too late for an individual, and the transition may be unsafe in several different ways. One story is about the consequences; the other is about an attempt to prevent those consequences in the absence of clear guarantees. Together, they are another reminder that we need to engage early with processes that are already entering many areas of our lives at great speed.