AI Isn't Management. Try Explaining That to Matthew Prince

At last we have created the Corporation That Eliminates Middle Managers from classic management text Don't Eliminate the Middle Managers!

Here’s another data point about the relationship between AI and organization. While rightwing politicians in other countries are fantasizing about DOGE-ing their civil service and replacing large swathes with AI, businesses are boasting about what they have already done.

Last Saturday, Matthew Prince, the CEO of Cloudflare wrote a self-congratulatory op-ed in the Wall Street Journal, suggesting his company was the only one in recorded business history to grow by 30% while laying off more than 20% of its employees. His message was that everyone needed to follow his example, by using AI to implement the True Wisdom of Revered Management Guru Peter Drucker.

It’s worth quoting Prince’s arguments at some length:

what we did is likely going to become the norm over the next year. This is a story about artificial intelligence, but executives and commentators are misunderstanding how it will disrupt business and who will be affected. To understand the issue, I went back to a book published in 1954, 20 years before I was born: Peter Drucker’s “The Practice of Management.” Drucker explores the different roles inside every business, which I would categorize as builders, sellers and measurers. Builders create products. Sellers sell those products. Measurers do everything else: internal audit, revenue recognition, finance, legal, compliance, middle management, operations and on and on. … builders aren’t going anywhere. … Sellers, too, are safe from extinction. … Measurers are also critical to a business, but different from the other two. …

Drucker argues that measuring business is important, but customers are earned through building and selling. The best businesses would maximize investment in those two functions. AI isn’t coming for builders or sellers, but it is coming for measurers. … AI won’t kill all jobs. But it will change every business. Ultimately, it will prove Drucker right. AI will allow us to better measure our organizations so the humans on our teams can focus on where they create and capture value: building and selling.

Matthew Prince is right about one big thing. You should absolutely read Peter Drucker’s The Practice of Management if you can. But while I presume that Prince is telling the truth about reading and re-reading Drucker, he doesn’t seem to have really understood what Drucker actually wrote.

Prince’s Peter Drucker wrote a book about how building and selling are crucial, and new technologies such as AI can be used to get rid of great swathes of middle management that don’t do more than measure what everyone else is doing, all for the benefit of the boss. Drucker’s Peter Drucker wrote a book about how all management involves measurement and planning and action on the basis of those measurements. Those measurements aren’t supposed to help the boss decide who to promote and fire, but to enable the managers themselves to do better. Drucker’s Peter Drucker is much less interested in keeping the accountants happy, or, for that matter, maximizing profits, than in increasing the capacity for managers to exercise human judgment, something that can’t be automated.

As I see it, Prince has badly misread an author whose understanding of business (at least in the book he cites; I haven’t read everything by Drucker) provides a radically different alternative to the standard Silicon Valley CEO narrative. If you take Drucker seriously, instead of whittling him into a peg to hang your boasts on, he offers a different, and potentially useful understanding of how AI ought be applied within the firm, and equally importantly, how it ought not

*******

Peter Drucker was an enormously interesting general purpose intellectual, even if he is rarely considered as such. He was more a theorist than an ordinary guru - a Harvard Business Review poll of business writers identified Drucker, James March and Herbert Simon as their favorite thinkers. That seems the right crowd to situate him in: all three were worldly philosophers in the broad sense, and their ideas about organization have a surprising amount in common.

Drucker was visibly influenced by Schumpeter, without Schumpeter’s arrant elitism, occasional sneaking regard for Fascism, or extraordinarily lively prose style; Drucker has some very good lines but is careful rather than exuberant. Schumpeter and Drucker also had quite different attitudes to the workings of the post-World War II economy. While Schumpeter deplored the strangulation of the hero-entrepreneur by management and bureaucracy, Drucker was prepared to make peace with the new dispensation, and celebrated managers as innovators. He used the term “manager” capaciously, not just to celebrate CEOs and their immediate underlings, nor even just middle-managers, but workplace foremen, who necessarily exercised managerial capacities and judgment as part of their responsibilities. If Druckerism is in part Schumpeterianism-lite, it is Schumpeterianism-lite nearly all the way down, focusing on cultivating initiative and self-actualization all the way through the rank-and-file of the firm.

Drucker believed that good management required giving a remarkable degree of leeway to managers. Instead of distinguishing between builders, sellers and measurers, he identified “measurement” as one of the five essential activities that all managers need to engage in, together with setting objectives, organizing, motivating and communicating, and developing people. The manager:

establishes measuring yardsticks, and there are few factors as important to the performance of the organization and of every man in it. He sees to it that each man in the organization has measurements available to him which are focused on the performance of the whole organization and which at the same time focus on the work of the individual and help him do it. He analyzes performance, appraises it and interprets it.

The problem of measurement is one of setting goals for oneself and keeping track of whether you are reaching them, and of communication. Everyone in the business needs to have some sense of what the business as a whole needs, and how his or her own activities meet those needs. Equally, everyone needs to exercise their own “judgment” (a term that Drucker employs repeatedly) as to how they carry out their particular role. Measurements are primarily supposed to orient the manager rather than judge and condemn him (or her, if we are writing in the 2020s rather than 1950s) from outside.

reports and procedures should be the tool of the man who fills them out. They must never themselves become the measure of his performance.

Measurement should absolutely not be used as a means of top down control:

I have not talked of ‘control’ at all; I have talked of measurement. This was intentional. For ‘control’ is an ambiguous word. It means the ability to direct oneself and one’s work. It can also mean domination of one person by another. Objectives are the basis of ‘control’ in the first sense; but they must never become the basis of ‘control’ in the second, for this would defeat their purpose.

Where measurement becomes an instrument of control, it can have disastrous consequences.

In [an unnamed] company a control section audits every one of the managerial units … The results of the audits do not go, however, to the managers audited. They go only to the president, who then calls in the managers to confront them with the audits of their operations. … the nickname the company’s managers have given the control section: ‘the president’s Gestapo.’ Indeed more and more managers are now running their units not to obtain the best performance but to obtain the best showing on the control section audits.

The quote about the Gestapo illustrates a more general point. Drucker believes that the pathological versions of top-down control-by-president-or-CEO have much in common with political dictatorship. Management, properly understood, is a mode of human autonomy, a liberal activity in the strong sense of the word. Squashing it into mere implementation is accordingly tyrannical.

Fundamental to Henry Ford’s misrule was a systematic, deliberate and conscious attempt to run the billion dollar business without managers. [italics in original] … Henry Ford’s concept was not even unique in industry. It was widely held in the early years of the century. He shared it, for instance, with one of his most distinguished contemporaries: Lenin.” … Above all, it seemed to make possible industrialization without management, in which the ‘owner,’ represented by the political dictatorship, would control all business decisions while business itself would employ only technicians.

Drucker was suspicious, on similar grounds, of Taylorist efforts to automate business activities so as to separate out planning from doing.

Planning and doing are separate parts of the same job; they are not separate jobs

By suggesting otherwise, Taylorism verges on authoritarianism:

the divorce of planning and doing was also part of the elite philosophy that swept the Western world in the generation between Nietzsche and World War I - the philosophy that has produced such monster offspring in our time. Taylor belongs with Sorel, Lenin and Pareto. … power must be grounded in moral responsibility; anything else is tyranny and usurpation

The role of the manager is not to plan what others do, or merely to implement the plans of others, but to exercise human judgment within a moral setting to integrate social activity:

The manager has the task of creating a true whole that is larger than the sum of its parts … One analogy is the conductor of a symphony orchestra, through whose effort, vision and leadership individual instrumental parts that are so much noise by themselves become the living whole of music. But the conductor has the composer’s score; he is only interpreter. The manager is both composer and conductor. … The task of creating a genuine whole also requires that the manager in every one of his acts consider simultaneously the performance and results of the enterprise as a whole and the diverse activities needed to achieve synchronized performance

Again, Drucker emphasizes repeatedly that this vision of managerial capacities does not apply to the top of the hierarchy but to everyone whose work is not completely technical.

It is no exaggeration to say that Drucker views management as a specific mode of realizing human potential. This is sweeping, romantic and occasionally a little corny. It also makes Drucker an enormously attractive writer and thinker in ways that many of his epigones are not. He views action and responsibility as going together, stressing both that business has a broader social role (profit is no more than a necessary condition to stay in operation, and business must pay attention to the common good), and that the task of the manager is to achieve a human synthesis of individual autonomous decision making with some reasonable understanding of the wellbeing of the firm and society.

*******

Matthew Prince’s goals seem notably different. Quoting the Wall Street Journal op-ed again:

Tireless, independent, efficient and available, AI systems can now measure an organization with a level of objective detail and precision that was previously impossible even for the best employees. … as CEO, I’ve never had better tools to measure exactly how the business is performing, including identifying our rising stars. The vast majority of those we laid off last week were measurers. We cut middle managers across the organization because AI allows us to have more direct reports per manager while still measuring and mentoring our teams effectively. … With fewer people needed for measuring, we can now invest more in people in the areas that drive growth

If you squinted with great vigor, you could perhaps just about interpret this as liberating managers from the tyranny of bureaucratic minutiae and establishing the conditions of autonomy that Drucker emphasizes. It does, after all, mention “mentoring” in passing. But those who are less charitable, such as myself, might emphasize the rather more explicit “as CEO, I’ve never had better tools to measure exactly how the business is performing, including identifying our rising stars.” That might in turn lead them to turn to a different analogy.

Deceased Business Intellectual: In my book I described Henry Ford’s Corporation Without Managers as a cautionary tale. Tech Company CEO: At long last, we are creating the Corporation Without Managers from classic business text Don't Create The Corporation Without Managers.

Matthew, Prince of Cloudflare, is deploying AI to better descry his realm as a whole, and to create a more efficient hierarchy. AI will allow him to reward those who the indicators say are valuable, while firing those whom he believes to be superfluous.

Do note a trick that appears and disappears very quickly in the text. Prince literally classifies everyone in his company who is not either building or selling as “measuring.” While he acknowledges that really good human measurers can be valuable, and presumably isn’t firing all of them, he claims that AI can do most measurers’ jobs more tirelessly, independently and efficiently than human beings. AI, then, doesn’t simply automate the carrying out of technical functions. It replaces managers who aren’t visibly building or selling stuff, doing their jobs better than human beings ever could.

None of this is in Drucker’s book, unless I’m missing something. Indeed, as far as I can see, Drucker believes the contrary. He insists that measurement fails its purpose if it is used as a top-down tool of assessment. He argues that the people who do not contribute visibly to the bottom line regularly play a crucial role, even if they sometimes can be mismanaged.

what the accountant lumps together as ‘overhead’ - the very term reeks of moral disapproval - contains the most productive resource, the managers, planners, designers, innovators. It may also, however, contain purely parasitical, if not destructive elements in the form of high-priced personnel needed only because of malorganization, poor spirit or confused objectives, that is, because of mismanagement.

And all of this stems from the notion that measurement is a chancy undertaking, and is especially chancy when it is separated from human judgment. Anticipating both Herbert Simon and a lot of the modern discussion of metrics, Drucker emphasizes how the “measurement used determines what one pays attention to. It makes things visible and tangible. The things included in the measurement become relevant; the things omitted are out of sight and out of mind.” While measurements are essential, they are highly imperfect. They can support managerial judgment but they cannot substitute for it.

to manage a business is to balance a variety of needs and goals. This requires judgment. … the attempt to replace judgment by formula is always irrational; all that can be done is to make judgment possible by narrowing its range and the available alternatives, giving it clear focus, a sound foundation in facts and reliable measurement of the effects and validities of actions and decisions. And this, by the very nature of business enterprise, requires multiple objectives

Such business judgment simply cannot be substituted by actually-existing AI. As Cosma and I have argued about other forms of organization (building on Ben Recht).

[optimization] simply does not provide any objective means of weighing the kinds of choices across non-commensurables that are essential to the bureaucratic process. Simon, Smithburg and Thompson (1958, 74) … note … “internal conflicts and contradictions among the ultimate objectives…” Such vexing choices regularly emerge in bureaucratic implementation and articulation, as well as in goal-setting at the top, making them fundamentally political problems, and this is just what optimization does not solve. As Recht (2023) bluntly says, “you can’t optimize a trade-off.” There is no consensus over how to optimize over multiple independent objectives in even moderately complex situations. Machine learning, with or without neural nets, does not offer magical solutions to this.

The obvious risk of identifying large groups of managers as mere “measurers,” and replacing them with AI, is that you later discover that they were doing a whole lot of irreplaceable tasks of judgment too, balancing across the different and complicated choices that Drucker, like Simon and his colleagues, identifies as crucial to managing. AI will certainly supplement measurement, and might usefully substitute for many aspects of it, if carefully implemented. However, it cannot reliably substitute for managerial judgment as Drucker describes it.

Perhaps Matthew Prince has just written a very sloppy op-ed; he and his immediate reports may possibly have correctly identified the tradeoffs in their hiring and firing decisions and calibrated them accordingly. Or perhaps not, in which case, he and his readers should consider Drucker’s pithy dictum that “It takes years to build a management team; but it can be destroyed in a short period of misrule.”

*******

Drucker’s actual philosophy of management provides an interesting alternative starting point for thinking about how AI ought be deployed in business. When Drucker wrote his book in the 1950s, both automation and “operations research” (optimization techniques which were an important ancestor of modern AI) were big deals. Drucker wanted managers to use information science to do their job better, but vehemently opposed claims that human judgment and human relationships could be magically automated away.

The Practice of Management expresses deep skepticism about “lurid ‘science fiction’ … about Automation,” invoking “visions of the “technocrat’s paradise, in which no human decision, no human responsibility, no human management is needed, and in which the push button run by its own ‘electronic brain’ produces and distributes abundant wealth.’ However valuable operations research was, it could not substitute for human management: “these are tools of information, and of information-processing, not of decision-making.” Failing to understand this would lead the manager, like “the Sorcerer’s Apprentice, [to] become the victim of his own bag of tricks.”

Back then too, there were grandiose claims about how This New Technology Will Fix Everything And Is Completely Inevitable, So Lie Back and Enjoy.

I have learned to be extremely sceptical of any prediction of imminent revolution or of sweeping changes in technology or business organization

and:

Now that we in the free world no longer accept the planners’ remedies as good for us, an attempt is being made to make us swallow the same nostrums under the pretext that they are inevitable.

Obviously, AI isn’t identical to early operations research or the Taylorist program to turn workers into optimizable machines. There are many things that modern AI can do more efficiently and reliably than past technologies, and indeed human beings. Equally, the differences between AI and past forms of automation are not nearly as marked as much of the current rhetoric might suggest.

Drucker suggests that

Procedures can work only where judgment is no longer required, that is, in the repetitive situation for whose handling the judgment has been already supplied and tested. Our civilization suffers from a superstitious belief in the magical effect of printed forms. And the superstition is most dangerous when it leads us into trying to handle the exceptional, non-routine situation by procedure.

Modern AI too is surrounded by superstitious beliefs about magical effects, and tends to fail when it encounters exceptional situations.

When street-level algorithms encounter a novel or marginal case, they execute their pre-trained classification boundary, potentially with erroneously high confidence.

The best bits of The Practice of Management describe a moral vision that subordinates technology to a particular understanding of human wellbeing. When Drucker argues that automation ought replace repetitive drudge work, he suggests that this will lead to more management rather than less.

the new technology will not render managers superfluous or replace them by mere technicians. On the contrary, it will demand many more managers. It will greatly extend the management area; many people now considered rank-and-file will have to become capable of doing management work. The great majority of technicians will have to be able to understand what management is and to see and think managerially.

The bigger point is that management is a moral, as well as an economic activity. It involves people developing and exercising their independent capacities of judgment towards the broader goals of the organization and the society that they live in. Drucker’s perspective thus has more in common with Pope Leo’s new encyclical than with Matthew Prince’s effusions. The similarities are unsurprising: Drucker was both a Christian (his book talks in passing about “Our Lord’s Parable of the Talents”) and a Christian Democrat.

A plausible reading of Drucker’s argument is the following: Businesses should pursue automation up to the point, and only up to the point that it supports the capacities of both traditional managers and rank-and-file to exercise individual human judgment in their work. That is not only in the best interests of the business, properly considered, but of the living community of individuals who constitute it. “Extending the management area” is not just an expected outcome of automation done right. It is also an appropriate moral goal, allowing work to move away from drudgery and toward the development and use of human capacities. Where automation change business organization so that it departs from this goal it hurts both the people and the organization itself.

This is not the only possible reading of Drucker’s book in the age of AI, but it seems to me to be highly congruent with his thought. It also has obvious implications for how we think about AI in the workplace. My version of Drucker might say the following. The current Silicon Valley approach to AI is profoundly problematic. It reinforces hierarchy and subordination to the vision of the CEO, rather than enabling individual judgment. It grossly simplifies many of the complex and multifaceted activities of management into “measurement” processes that AI can purportedly replace. It subordinates the moral and developmental aspects of business activity to abstract measurements of efficiency. If someone were actually to write an op-ed that applied Drucker’s ideas to AI, and its associated business model, they would say very different things than Matthew Prince.