Hello everyone. Today we are continuing our theme — the conversation about AI, about people, about work and about where all of this is heading together with AI. I want to share a story. A couple of years ago, at a management meeting at my own job, where I was one of the participants, someone put forward a hypothesis. I remember it very well, and it went roughly like this: why don't people understand how beneficial using AI is? After all, with AI we will spend more time with our families and our friends, and more time on what we actually care about.
I remember my own answer almost word for word, because I disagreed and suggested that with AI we would all work even longer. And AI would certainly not automatically become the tool that lets us spend more time where we want to be. My point was not that AI would become such an indispensable part of our work at the time; what I meant was that the freed-up time would fill very quickly with new demands and new expectations. I am not claiming that every single person will work longer because of AI.
What I want to highlight is that the system we are in very often works that way by default. If more can be done in an hour, more starts to be expected of us. That is what I want to talk about today.
If AI saves time, who ends up with the time it saved? And does AI actually make a person freer anywhere? Or does the decision about that freedom sit in an entirely different loop, one that does not depend on technology at all? The time we save — where does it actually go? You often hear the answer in general terms, because when we talk about AI we talk about saved time, and on the other hand we talk about becoming more productive. So if we produced a letter very quickly, or wrote some code, or made a presentation, does that mean we became more productive, more efficient?
Did the whole chain of work after us become more efficient? So: time, freedom, and how saved time converts into output and productivity. Let us start with the first line. AI saves time. What happens to that time? The unit of output we generate becomes cheaper. Why does it become cheaper? Because it very often becomes many times, tens of times faster. But, as I said above, we very rarely stop at the previous volume. We simply start producing more, we start producing faster. We used to write a text by hand in fifteen minutes; now we can dictate it in a minute and a half, and so in five minutes we simply dictate far more text than we could have written before.
With AI, generating most documents becomes instant, and the freed-up time, notice, is very often loaded with a new task. The technology shortens one activity in terms of time, but the system around it fills that freed time very quickly. I came across a rather interesting study from the University of California, Berkeley on exactly this question. The researchers spent a long time — around a year, or eight months — observing one technology company and where the employees' saved time was going.
They ran around forty interviews. More than two hundred people inside that company took part in the study. And what did they highlight? The researchers found the opposite side of acceleration. Employees worked faster, but in parallel they took on a larger and larger volume of tasks. They ran several parallel processes at once and very often stayed after work, without any manager asking them to. In effect, AI creates this layer of work where you can always do it better, go deeper, add something, check it again.
And I think this connects well with what we talked about last time — the widening of the boundaries of a profession. On the one hand we do more and more tasks; on the other hand we start moving into the territory of other professions, into adjacent specialisations. So there is a whole complex here, both about widening and about work changing under the influence of artificial intelligence. Perhaps this really proves nothing on its own, but the mechanism is very clear: our freed-up time fills up with new expectations almost immediately.
I have worked in human resources for more than twenty years and I see it in my own work. I keep finding more and more cross-functional links with adjacent functions. My job is acquiring an ever wider range, because I can create more and more cross-functional connections with the help of AI. At the same time I speed up considerably on other tasks, but my work does not get smaller, and in any global sense it does not get faster — even though, as I said, I use artificial intelligence very seriously and very deeply on a large number of tasks.
In general I think that working less is an entirely different loop, and technology will certainly not make that decision for us, because that decision rests with people. It rests on a person's own understanding of what they produce, of labour, of their profession. That decision rests with society and it rests with businesses. There is a separate and very large question of whether competition in the market, the way a company is built and our own expectations allow us to stop at all.
And if we go by the views of the people currently running technology companies, then apparently we are all entering and sustaining a new race every day, and simply climbing onto another layer. Now, further. If we have assumed and accepted that the time AI saves does not make us freer — or at least that technology is not what decides it — then where does that time convert to, where does it end up? And here comes a logical assumption: that AI raises our productivity, that it makes us more efficient, that we begin to fill the saved time with new output which we create, per unit of time, some number of times greater than before.
And that is true. It would of course be a complete mistake to say that AI does not raise output — it does. There is a great deal on this today, and I think everyone has personal examples. There are plenty of interesting studies and evidence. But here I want to highlight a slightly different side of the question. AI raises our output, but does the whole job get better? And how do we assess and measure that output? How do we assess and measure that productivity? Does the speed of the first draft mean the productivity of the entire process?
Does saved time always convert into a productive and efficient process? For example, in a few minutes I consolidate analytical documents from my team of ten or fifteen people. Then another person spends several hours trying to understand what was lost in that consolidation, what the logic was, why certain facts or hypotheses do not line up with one another. Where exactly does the efficiency appear here: in the fact that I consolidated quickly, or in how long that other, third person spends trying to make sense of the consolidated information?
I run into this very often at work these days, and lately I fairly regularly hear my team decide that it would be faster to do it by hand. Or the request is framed from the start with specific parameters for the output, because more and more often we encounter a situation where a low-quality output from a generated document raises our downstream work exponentially. And our saved time fills up very quickly with other people's labour. As a rule we count the minutes of generation, but we do not count the hours spent framing the task nearly as often.
We do not often count how much time another person spends checking and working with the result we handed them. We do not count the time for subsequent approval, implementation and the repeat cycle that is very often necessary — running a particular result through one model or another again. Researchers from Stanford University, together with New York University and MIT, have described what amounts to a new layer of labour created by generative artificial intelligence.
They say that a genuinely new layer of work is forming, one that includes collective checking, re-checking, experimentation, coordination, implementation, subsequent rollout of changes and subsequent verification. So it turns out that AI does not simply solve some question quickly — it effectively creates new work around itself, work that very often stays invisible. And notice how often our saved time simply creates new work for other people, work they spend on understanding, correcting and dealing with the consequences.
By the way, I would be interested — please share in the comments: with artificial intelligence, have you personally started working more, or started doing different tasks?
How exactly has your work changed, and where does the time you save go? There is a second level here as well. Is AI productive, and how does that depend on readiness or unreadiness to change processes inside the company itself, inside the business, or inside any other process? AI never layers onto empty ground; it layers onto current processes. If we do not change the processes, then in effect we multiply the existing errors, or multiply historical habits that may already have been inefficient and were not delivering the value we needed.
There have been several videos on Alexander's channel about the combination of AI and the automation of real business processes. The key point is that before adding AI, let us look at whether the process is needed at all. And let us not automate, or add artificial intelligence to, processes that have not been needed for a long time, that are inefficient and that should be the first thing we drop. Here, of course, I observe from various sides what is genuinely a super-fear these days, fed by the information space around us.
The fear of falling behind, the fear of not automating something, the fear of not using AI in one process, another, a third. Even the fear of not appearing to show enough AI. In many companies today AI is declared a strategic priority, and everyone rushes to use AI wherever possible, without thinking about where the saved time then converts to. Does this generate new work for other people? And does that use, at that moment, really deliver the productivity and the efficiency everyone is chasing and wants to talk about?
As an interesting confirmation of these thoughts — there is in fact a mass of research — I also came across a recent study by the international management consultancy McKinsey.
They published a study of seven hundred and fifty employees and managers who use AI. What is interesting is that more than seventy per cent felt personally ready to work with AI, but only around twenty-seven per cent of managers considered their organisation and processes ready for change. And among the companies that had deployed AI, thirty-two per cent of leaders reported business value where the workflows had been reviewed and rebuilt, against six per cent where nothing had been changed.
So in essence what these studies confirm is a fairly sound logic: the key task is not using AI, the key task is understanding where using AI delivers real business value, and what business value the company actually needs today. More and more real cases are appearing where companies do not abandon AI altogether but stop a specific tool, a specific solution. I want to give three examples that, it seems to me, illuminate this stage of stopping from different angles.
This is Starbucks, first and foremost an international coffee-shop chain. A couple of months ago they stopped a very large project that used computer vision to identify various products on the shelves — milk, syrups — to run inventory and perform a range of other tasks with technology. What it all led to was that staff had to rescan endlessly and re-verify the result, because the tool confused similar items, misread products and so on. To be clear again: I am not saying that Starbucks abandoned its AI plans. We are talking about stopping one specific scenario, because after nine months of use the company understood that going further was not efficient.
Another interesting example comes from one of Australia's largest banks, Commonwealth Bank of Australia. The bank launched a voice AI bot, and after it was launched the bank announced that forty-five contact-centre staff would be cut — but then reversed that decision fairly quickly and apologised to the employees, because it acknowledged that a number of business factors had not been taken into account when the bot was launched, and as a result the load on the contact centre turned out to be more than the organisation could handle.
The third example is Ford, the American carmaker. It is a very telling one. Over the past three years the company has brought back and hired more than three hundred and fifty experienced engineers, who spot and find the various gaps in automation, who train young engineers and pass their knowledge on. Because the path the carmaker chose — handing those tasks to various automated systems — did not work, and the result they now see from bringing back human expertise was not something they could achieve that way either.
Look, there are surely a great many examples like these, and these are not particularly special ones. They seem telling to me because they show, from different angles, how companies can stop a specific tool — as in the case of Starbucks — or reverse a premature staffing decision involving redundancies. We hear, see and read quite a lot about that now. Or else there is the deliberate return of human expertise to the process, as with Ford. And this is not about abandoning AI; it is about abandoning the belief that using AI always equals productivity, and that human expertise is not needed.
It also highlights something very interesting for all of us: dropping a scenario, saying «stop» to AI, does not mean defeat. Very often it means a hypothesis was tested, an experiment was run, and that is movement forward rather than backward. Because in truth nobody taught us how to handle this; nobody taught us from childhood how to work in an environment like this. And we now talk a great deal about needing to write a good prompt, about needing to treat the result consciously, about needing to keep control of agents, about not giving artificial intelligence the right to make decisions a person is answerable for.
But so many of these things have extremely fine lines; it is an endless balancing act. And where is any of this taught systematically? The question is not even about universities. Last time, for instance, I talked about how even specialised courses at technology universities may have no AI track at all today. What this is really about is requirements, new qualities of thinking, new competencies that have to be formed — and that are almost certainly layered in from childhood, through parents, through nursery, through school, through university.
It requires an entirely different methodology even for forming those competencies. But what do we do today? We all understand this, and yet here we are, now, today.
And whether AI works or does not work, whether it is productive or not, we still have to make these decisions today. I want to propose asking questions. To solve this problem and answer this question — how to check quickly — I suggest using very simple questions. I have settled on five for myself. There are surely many more that could be defined. Perhaps the frame could be narrowed. But if the work of an entire process and of other people depends on our result, it is very important always to take this step consciously and to understand what we are investigating and launching, and why.
So, question one: which concrete outcome for the business or for the customer should get better? What exactly will change? Speed will change, quality will change, perhaps the safety of some process will change. Let us always define clearly what exactly we will get, what result. Question two: what old work will disappear? If, as we said today, using generative AI effectively creates a new layer of work, a particular layer of mental operations that did not exist in this volume before — then at the very least we should understand what AI will replace.
If the previous documents and processes remain fully in place, then most likely we are simply adding one more layer on top of another added layer. It becomes an endless staircase. Question three: did the full time shrink? Not only the time to the first output, not only to the first generation, but the time afterwards. The time we spend on what we do with that document: how long we check it, how long we spend correcting it, how long we need to get it approved, knowing that the initial generation was done by artificial intelligence.
So let us look at and assess the full time. Question four: who understands the result, who is answerable for it, and at what result do we press that «stop» button? As in the examples we discussed today, every process has an owner and every process has a person who answers for the result. But a competent, timely stop of that process is just as much a result. We have already said that very often this is not defeat but movement onward. We tested a hypothesis, we ran an experiment. It did not work out — excellent, we do not use it.
That in itself is a good outcome. And question five. It is the question we started with: where will the freed-up time go? We will reduce people's workload. We may work, and complete tasks, with fewer people. We will produce more units of output. We will go and rest, or we will take on new tasks. One way or another, that saved time converts into something. And it is very important to remember this, because I also see a trend right now where we do one task faster, but if the remaining tasks do not fill the freed capacity — and, as we discussed, freed capacity fills up very quickly — then perhaps the old tasks simply swell, and we are in fact living in an illusion of efficiency.
Let us ask these questions every time. And there are probably many more questions than these. There are questions that let you get to the root and find answers faster. But I believe this is the right trend and the right current, so that we do not deceive ourselves and do not deceive the place where we are. In essence I am not changing the answer or the assumption I set out several years ago.
I genuinely do not think AI makes us free, and working less is definitely a different loop. It is decided by the person, it is decided by the manager, it is decided by the system we work in. On the other hand, I think we have managed to show today that speed does not automatically give anyone the right to call themselves more productive. Speed does not automatically give anyone the right to claim business value. If the old work does not disappear, then perhaps we are simply doing more of the old work and endlessly multiplying whatever errors or inaccuracies were already in it.
AI is not something to be afraid of — but rushing into super-productivity is often not worth it either. Let us ask questions.