In a world increasingly reliant on computer programs and applications, a group of Brazilian researchers spent more than four years studying how the software industry operates, seeking to explain its dynamics and how it is organized. The result was a theory detailing the software production process in Brazil and around the world. Understanding this structure, experts argue, is essential to ensuring continuous and reliable delivery of these systems, used everywhere from hospitals and air traffic control centers to banks, social networks, and streaming services.
“One of the main goals for those who study the field of software engineering is to understand how people come together and coordinate collectively in the task of developing computer programs,” says software engineer Leonardo Leite. An employee at the Federal Data Processing Service (SERPRO), Leite carried out the investigation during his PhD at the Institute of Mathematics and Statistics of the University of São Paulo (IME-USP). His thesis, titled “A grounded theory of organizational structures for development and infrastructure professionals in software-producing organizations,” won the software engineering category of a thesis competition organized by the Brazilian Computing Society in 2023.
DevOps, short for development and operations, is a widely adopted practice among Big Tech companies, software firms, and many other organizations. In this collaborative approach, software is created by developers and closely monitored by infrastructure or operations teams, which are responsible for keeping the software running smoothly once deployed, ensuring availability, reliability, and performance.
“The synergy between the two teams allows everything to function properly, regardless of demand at a given moment,” says Leite. “At the same time, it facilitates daily software updates that users barely realize are happening. This happens with Netflix, the Brazilian digital bank Nubank, Amazon, and many other apps that we use on a daily basis.”
The methodology applied in the research, known as grounded theory, is designed for developing theoretical frameworks based on data collected in the field rather than predefined hypotheses, which is more common in science. With this approach, theories are constructed step by step from empirical evidence.
The first stage of the study, funded by FAPESP, consisted of a broad review of the literature on DevOps. After analyzing 50 papers on the topic, Leite and his colleagues published an article in the scientific journal ACM Computing Surveys in 2019 describing the challenges faced in the study of DevOps and offering recommendations to industry professionals and academics.
“In that article, which has since been widely cited, we presented a concise definition of DevOps, a controversial term for which there was no consensual understanding,” says the researcher. “For us, DevOps is a collaborative, multidisciplinary effort within an organization aimed at automating the continuous delivery of new software versions, while ensuring their correctness and reliability.”
Anonymous interviews
The next step in the research was a qualitative study of 75 professionals from 59 Brazilian and international companies of varying sizes. To encourage professionals to speak openly about the problems faced at their companies, Leite chose to use anonymous interviews. From this data, he identified four organizational models in the software industry: silos, collaboration, unified teams, and platform mediation (see infographic).
In the silo model, the development and operations teams are separated, often even working in different buildings. They communicate with each other via email or other digital channels. This approach is most prevalent in large corporations. “This is the oldest organizational model. In Brazil, it is still common,” says IME-USP computer scientist Fabio Kon, one of Leite’s PhD supervisors.
The second model, by contrast, brings developers and operational staff together in unified teams or single departments, working in the same room or virtual environment. Sometimes, the same professionals even take on both roles. “It’s a step forward from the silo model because it makes communication more efficient,” says Kon. It tends to be adopted primarily by small companies, but Amazon also uses it.
The third organizational design identified by Leite—collaboration—was most common at medium-sized companies. “If a company has 200 developers,” Kon illustrates, “it might form 20 teams of 10. An additional operations team of 10 provides daily support to the developers.” The difference between this structure and the silo model is that communication is more direct and efficient.
“The operations staff visit each development team on a daily basis to assist them with their needs,” says Leite. At very large companies, the researcher explains, it is difficult for the centralized operations team to support thousands of developers. “Therefore, when a larger company like Google adopts this model, some adaptations are necessary,” says Leite.
Finally, the platform model, in which the software production process, including communication between the development and operations teams, has a high degree of automation and takes place through a platform or application programming interface (API)—a system that enables communication between two machines. It is the most efficient approach, although it is not recommended for smaller companies. “The four models coexist in the industry. None is universally more suitable than the other. The choice depends on several factors. The more complex the operation, the greater the tendency to migrate to a platform model,” says Kon.
“The grounded theory method was applied properly,” says computer scientist Elisa Yumi Nakagawa of USP’s São Carlos Institute of Mathematical and Computer Sciences (ICMC). “Although the number of people involved in the study might seem small compared to the total workforce, 75 participants was enough to reach theoretical saturation. Even if more people were interviewed, the theory would likely remain the same.” In grounded theory, the theory is formulated as each interview is analyzed. It tends to undergo more changes during analysis of the earlier interviews, but then becomes more stable as the number of interviews increases.
The fieldwork consisted of interviews with 75 professionals
According to Leite, previous attempts to create such taxonomies did not follow scientific methodologies and lacked robust datasets. “Our taxonomy reflects situations actually encountered in the field. It is not a set of ideal types that could theoretically be achieved, but a theory based on data representative of observable reality.”
The research led to the publication of two articles in international journals. The first, published in Information and Software Technology in 2021, described the four software production organization models observed in the industry. The second, published in IEEE Transactions on Software Engineering in 2023, focused on the advantages and disadvantages of each.
Later, the team collaborated with researchers from the Polytechnic University of Madrid (UPM), who conducted a similar study in Spain and arrived at comparable results. “We worked together to harmonize our taxonomies, in an effort to establish a unified model based on both perspectives,” says Leite.
Practical applications
Experts interviewed for this report believe that Leite’s theory has important practical applications in the software industry. “It is especially important for organizations grappling with the challenges of adopting DevOps,” points out Paulo Meirelles, a software development technologist at IME-USP’s Computer Science Department and Leite’s PhD supervisor. “The taxonomy of the four approaches helps companies identify where they stand in their DevOps transformation process.” It also allows them to understand the benefits and limitations of each model.”
Computer scientist Gustavo Pinto of the Federal University of Pará (UFPA), who collaborated in one phase of the study, agrees. “DevOps is not a new topic, but the study exposed a latent problem in the structure,” he explains. “Many papers on DevOps are inconsistent and use different terms for the same practices. For readers, this often creates more confusion than clarity.” According to the researcher, the taxonomy identified in the research can serve as a bridge between academia and industry, guiding not only the structuring of teams but also tool development.
Alexandre Freire Kawakami, director of engineering, privacy, security, and user protection at a Big Tech company based in São Paulo, believes Leite’s theory could accelerate the rate at which many companies learn about and adopt DevOps. “The theory and the four organizational models make sense, as does the definition of DevOps,” he says. “Over the course of my career, I have worked in all four structures.”
To expand the study’s reach, Leite, Meirelles, and Kon published a book in 2025 titled How DevOps Works: Organizing People, from Silos to Platform Teams (Casa do Código), which was a finalist for the Academic Jabuti Award. “The book translates the research results into more accessible language for industry professionals,” says software engineer Eduardo Martins Guerra of the Free University of Bozen-Bolzano, Italy, who was not involved in the project. “Our goal was to build a bridge between academia and industry so that our research can have a real-world impact,” adds Leite. An English version of the book was published this year.
The story above was published with the title “Deciphering the software industry” in issue 355 of September/2025.
Project
INCT 2014: The internet of the future (n° 14/50937-1); Grant Mechanism Thematic Project; Principal Investigator Fabio Kon (USP); Investment R$2,049,398.08.
Scientific articles
LEITE, L. et al. A survey of DevOps concepts and challenges. ACM Computing Surveys. Nov. 14, 2019.
LEITE, L. et al. The organization of software teams in the quest for continuous delivery: A grounded theory approach. Information and Software Technology. Vol. 139. Nov. 2021.
LEITE, L. et al. A theory of organizational structures for development and infrastructure professionals. IEEE Transactions on Software Engineering. Vol. 49, no. 4. Apr. 2023.
DÍAZ, J. et al. Harmonizing DevOps taxonomies — A grounded theory study. Journal of Systems and Software. Vol. 208. Feb. 2024.
Book
LEITE, L. et al. Como se faz DevOps: Organizando pessoas, dos silos aos times de plataforma. Casa do Código. 2024.
Republish
