Efficiency alone does not generate value, and companies need to rethink strategy to scale AI

During the event "Efficiency doesn't pay the bills: the paradox of AI in organizations", promoted by Squadra Digital in partnership with Sensedia and MIT Sloan Management Review Brasil, executives Romulo Cioffi, Chief AI and Innovation Officer at Squadra, Lucas Tempestini, Head of Marketing at Sensedia, and Douglas Souza, CEO of MIT SMR Brasil, argued that the market is undergoing a change in how the success of investments in artificial intelligence is measured. After the race to experiment with generative models, the challenge is now to transform technology into real business impact.

The main conclusion of the study presented during the meeting is that operational efficiency, although important, is no longer sufficient to justify investments in AI. Without a strategy based on architecture, data integration, and governance, a large part of the initiatives remains restricted to pilot projects, without consistent generation of financial value.

According to the data shared at the event, 66% of executives state that generative AI holds a priority position in companies. However, only 7% of organizations have integrated technical capacity to operate these initiatives at scale, highlighting a disconnect between discourse and execution. Furthermore, only a small portion of projects manage to move beyond the experimental phase and effectively reach production.

"The market is still very focused on the tool. The competitive advantage, however, lies in the ability to transform intelligence into operation and decision-making," the participants summarized.

The problem isn't creating agents

One of the main points of the debate was the rapid popularization of AI agents. For the experts, many organizations have started to measure maturity by the number of agents developed, when the discussion should begin much earlier.

The assessment is that creating agents without understanding the company's processes tends only to reproduce technological silos and increase operational complexity. Before that, it is necessary to design a decision architecture capable of defining the role of each agent, its interaction with people, data, and corporate systems.

The proposal presented during the panel is to view the company as an "agentic organization", where intelligent agents cooperate with each other and with human professionals, always guided by business objectives and not just by task automation.

Governance ceases to be a barrier and starts to accelerate innovation

Another consensus among the participants is that governance should not be seen as a mechanism to limit AI projects, but as the structure that allows innovation to be scaled safely.

The executives argued that agents need to be treated as new users within the corporate architecture, subject to the same authentication, authorization, monitoring, and access control policies already applied to employees and traditional applications.

Recent cases of inappropriate use of generative models were cited to illustrate the risks of deployments without controls. Among them, chatbots performing functions for which they were not designed and systems accessing information beyond the intended scope, situations that reinforce the need for mechanisms capable of clearly defining which data and operations each agent can use.

In this context, the evolution of MCPs (Model Context Protocol) also entered the discussion. Just as happened with the adoption of APIs in recent years, experts argue that these components need to incorporate their own management, scope control, and traceability layers to prevent agents from having indiscriminate access to corporate assets.

Costs also enter the equation

Another lesson highlighted by the participants is that scaling AI does not mean concentrating all applications in a single language model.

According to the experts, different models present distinct performance and costs for each type of task. The trend is for organizations to adopt hybrid architectures, combining proprietary models, open-source solutions, and smaller internally developed models to balance performance, availability, and operational cost.

The discussion refers to the movement experienced during the migration to cloud computing, when companies realized that not all workloads should remain in a single environment.

AI increases productivity but requires cultural transformation

The panel also addressed the impacts of AI on the work of technology teams.

Instead of replacing professionals, the expectation is that developers will take on more strategic roles, acting as curators of the responses produced by the models, validating results, and concentrating efforts on higher value-added activities. However, the change depends on continuous training and a cultural transformation within organizations.

The participants highlighted that artificial intelligence significantly accelerates software development but also increases the generation of new demands. The greater the delivery capacity, the larger the backlog of application evolution tends to be, making the need for team adaptation permanent.

More than a race to adopt intelligent agents, the message left by the meeting is that the success of AI will depend on companies' ability to build a solid foundation of architecture, data, and governance. Without these elements, technology may even generate efficiency gains, but it will hardly produce the financial and strategic impact expected by organizations.

Available at: TI INSIDE

Sparked your curiosity and want to analyze the study data in full? Access and download it now.