Software Development as a Discovery Procedure

Gene Callahan
NYU Tandon School of Engineering

Nobel-Prize-winning economist F.A. Hayek was one of the most significant social theorists of the 20th century. Some despise him for his "neo-liberalism," and the comfort he might have given to the Pinochet regime in Chile. Others love him for his support of free markets and his defense of a liberal social order. But love him or hate him, no one should doubt his significance as a thinker: he did important work on the theory of the business cycle, on monetary theory, on the theory of capital, on the informational role of market prices, on the nature of complex phenomena, and on the importance of group selection in evolution.

In this essay, I wish to argue that, whatever one thinks of Hayek's politics, his work has important insights to offer those advancing Lean / Agile / DevOps ideas for IT. Here I will focus on his paper "Competition as a Discovery Procedure," and note how similar Hayek's vision for the role of competition in the market is to the Agile understanding of the importance of the "development" part of the phrase "software development."

That essay of Hayek's was written in response to the model of "perfect competition" that had come to dominate economics in the middle of the 20th century. In that model, "competition" meant a state of affairs in which each market participant already knew every relevant detail about the market in which they participated, and thus simply "accepted" a price that, somehow, mysteriously emerged from the "given data" of their market. In such a situation no actual competition, as it is commonly understood, really occurs: every "competitor" already knows what product to offer, what price to charge, and simply passively accepts their situation as it stands.

Similarly, the waterfall model of software development simply assumes that what has to be discovered, in the process of software development, is already fully known at the start of the process. Instead of correctly understanding development as a process through which the analysts, coders, testers, documenters, and users come to a mutual understanding of what the software should really be like, the waterfall model posits that certain experts can fully envision what the final product should be, right at the start of the process. "Software development" then consists of these experts drawing up a document analogous to one of the "five-year plans" of the Soviet Union, detailing how all of the other "participants" should work, according to the experts' plan. No further input is needed as far as what the software being "developed" should actually do. But in reality, as Eric Evans notes:

"When we set out to write software, we never know enough. Knowledge on the project is fragmented, scattered among many people and documents, and it's mixed with other information so that we don't even know which bits of knowledge we really need. Domains that seem less technically daunting can be deceiving: we don't realize how much we don't know. This ignorance leads us to make false assumptions." (Evans, p. 15)

Hayek, describing the dependence of economists on the perfect competition model, admits:

"It is difficult to defend economists against the charge that for some 40 or 50 years they have been discussing competition on assumptions that, if they were true of the real world, would make it wholly uninteresting and useless. If anyone really knew all about what economic theory calls the data, competition would indeed be a very wasteful method of securing adjustment to these facts." (Hayek, 179)

He goes on to write:

"In sports or in examinations, no less than in the world of government contracts or prizes for poetry, it would clearly be pointless to arrange for competition, if we were certain beforehand who would do best... I propose to consider competition as a procedure for the discovery of such facts as, without resort to it, would not be known to anyone..." (Hayek, 179)

This, I suggest, is quite analogous to software development: it would be pointless to engage in such a time-consuming, mentally challenging activity if we knew in advance what software "would do best." We engage in software development to discover "such facts as, without resort to it, would not be known to anyone." It is only when we put our interface in front of real users that we find out if it really is "intuitive." It is only when we confront our theoretical calculations with the real data that we know if we got them right. It is only when we put our database out to meet real loads that we can tell if its performance is adequate. We can only tell if our CDN design meets our goals when it actually has to deliver content. None of this means that we should not plan as much as possible, in advance, to make sure our software is up to snuff, just that how much is possible is quite limited.

Hayek highlights the true value of competition in the following passage:

"[C]ompetition is valuable only because, and so far as, its results are unpredictable and on the whole different from those which anyone has, or could have, deliberately aimed at... We do not know the facts we hope to discover means of competition, we can never ascertain how effective it has been discovering those facts that might be discovered... The peculiarity of competition -- which it has in common with scientific method -- is that its performance cannot be tested in particular instances where it is significant... The advantages of accepted scientific procedures can never be proveed scientifically, but only demonstrated by the common experience that, on the whole, they are better adapted to delivering the goods than alternative approaches." (Hayek, 180)

Bjarne Stroustrup, the creator of C++, has very similar things to say about programming:

"When we start, we rarely know the problem well. We often think we do... but we don't. Only a combination of thinking about the problem (analysis) and experimentation (design and implementation) gives us the solid understanding that we need to write a good program... It is rare to find that we had anticipated everything when we analyzed the problem and made the initial design. We should take advantage of the feedback that writing code and testing give us" (Stroustrup, 178).

Given that competition is a discovery procedure, and thus we can't ever predict, with certainty, the results of market competition, Hayek considers what sort of predictions economists can make, if any? After all, if economics is a science, we expect it to say at least something about the course of events. Hayek concludes that:

"[The theory of the market's] capacity to predict is necessarily limited to predicting the kind of pattern, or the abstract character of the order that will form itself, but does not extend to the prediction of particular facts." (Hayek, 181)

Similarly, in software development, although we can't anticipate in advance exactly what lines of code will be needed... or development would be done!... we can anticipate that good software will exhibit certain patterns. And thus we see Hayek anticipating the "pattern language" approach to software development that was imported from the architectural works of Christopher Alexander into software development.

Let us turn aside from contemplating the market order, upon which Hayek focuses most of his attention, and consider the other order Hayek mentions: science. Although any scientific enterprise involves planning, we cannot possibly plan out in advance what discoveries we will make in the course of some scientific research: if we knew those, we would have already discovered them, and our research would be done: we would just be writing up the results. But that is precisely what the waterfall model supposes: we already know what the software in question must do: development is complete, and all that remains is to turn the requirements into an executable program: essentially, just "writing up the results." This approach actually blocks the process of discovery, as it leaves no room for the developers or the users to achieve new realizations in the process of turning the blueprint into working code, realizations that would expose the "specs," the master plan, as being based upon false hypotheses.

One aspect of recognizing an order as a discovery procedure is the implication that where in an organization the most relevant discoveries will be made is also not predictable in advance. Many scientific discoveries have been made because a lab assistant failed to follow some accepted procedure, or noticed something her "betters" had missed. And many successful market innovations arose at the level of the factory floor or the sales visit, and not in the executive suite.

The waterfall model assumes that every insight about the proper form of the final software product will come from the "analysts," and that it is the job of "the workers," such as programmers, to simply turn those insights into executable code. In this respect, the waterfall model has much in common with "Taylorism," the blueprint for mass production pioneered by Frederick Taylor around the turn of the last century. As Jerry Muller describes it:

"Taylorism was based on trying to replace the implicit knowledge of the workmen with mass production methods developed, planned, monitored, and controlled by managers. 'Under scientific management,' [Taylor] wrote, 'the managers assume... the burden of gathering together all the traditional knowledge which in the past has been possessed by the workmen and then of classifying, tabulating, and reducing this knowledge to rules, laws, formulae... Thus all of the planning which under the old system was done by the workmen, must of necessity under the new system be done by management in accordance with the laws of science.' (Muller, pp.32-33)

But Taylorism and similar top-down approaches proved inadequate in manufacturing, as demonstrated by the triumph of the Toyota Production System, just as top-down planning failed in the Soviet Union, and just as it does in science. Perhaps ther most important piece of wisdom contained in the Lean / Agile / DevOps movement is that the waterfall model of software development fails for very similar reasons.

Once we recognize software development is a discovery procedure, it should prove useful to categorize some of the features of a program that are most likely to be discovered in the actual process of development, rather than having been perfectly anticipated in our initial analysis of our users' requirements. What I offer here is only intended as an initial cut at what surely is a much more extensive list that could be developed. With that caveat in mind, in the process of actually developing software, here are some likely areas where our initial analysis will fall short of the mark:

In conclusion, I suggest that recognizing the similarity between the two "discovery procedures" that Hayek identified, market competition and science, and a third such procedure, software development, ought to open fruitful avenues for future research in the Lean / Agile / DevOps space. In particular, theorists and practioners of modern software development methods might look to the work of Hayek, and his leading students, such as Israel Kirzner and Nobel-Prize winning economist Vernon Smith, for further insights concerning the dynamic nature of the software development process.


Domain-Driver Design: Tackling Complexity in the Heart of Software, Eric Evans, Addison-Wesley, Upper Saddle River (New Jersey), 2004.

"Competition as a Discovery Procedure," in New Studies in Philosophy, Politics, Economics and the History of Ideas, F.A. Hayek, University of Chicago Press, Chicago, 1978.

Programming: Principles and Practice Using C++, Bjarne Stroustrup, Addison-Wesley, Upper Saddle River (New Jersey), 2014.

The Tyranny of Metrics, Jerry Z. Muller, Princeton University Press, Princeton, 2018.

The Black Swan, Nassim Nicholas Taleb, Random House, New York, 2010.