AI as a power structure: from discourse to executive capability
Alcebíades Araújo, an artificial intelligence specialist from SQUADRA, shares his insights on the new report from TEC.Institute with MIT.
The report The GenAI Map in Brazil, from Strategy to Implementation, developed in partnership by TEC.Institute and MIT, starts from a simple and hard-to-ignore diagnosis: artificial intelligence is no longer a promise but has become strategic infrastructure. Instead of asking if we should adopt AI, the question becomes under what conditions and with what governance we will adopt it.
The message is clear: "technology without institutional capacity becomes a cost; with capacity, it becomes a competitive, regulatory, and social advantage."
The shift is evident in the disconnect between conversation and execution. Most organizations are already discussing GenAI in decision-making forums, but a much smaller fraction has a formal strategy in place. The result is a landscape marked by high awareness and incipient implementation. This gap is not technical. It lies in the capacity to decide, prioritize, measure, and integrate. Without explicit choices of relevant problems, without business-linked metrics, without data architecture, and without clear responsibilities, AI solidifies into perpetual experimentation.
The report proposes a reading path that avoids superficial technophilia and organizes maturity across five fronts that, when addressed together, transform technology into a management agenda.
1. Strategy and Leadership
Without unequivocal direction and sponsorship that spans across departments, AI becomes just a collection of disconnected pilots.
2. Governance and Responsibilities
The institutional design that defines roles, usage regulations, risk controls, and interface with compliance.
3. Ethics and Risks
This is not about stating generic principles but about making them operate as clear guidelines through model audits, human review mechanisms, and incident response plans.
4. Training and Technical Culture
With differentiated paths for leaders and teams, and direct confrontation of Shadow AI, or the informal and ungoverned use of tools.
5. Implementation and Integration
Reliable data, APIs and events, observability, versioning, and value measurement coupled with the process, not alongside it.
When we look at the maturity distribution, we see a country that observes and explores more than it builds and orchestrates. There are many tests, little value capture. This is reflected in integration: a large part is in pilots, a smaller portion reaches partial integration, and a minority achieves full integration. And it is in this minority that cases of concrete return are concentrated.
The reading is consistent with the J-curve dynamic: the initial phase tends to combine investments, frustration, and noise; results appear when the organization restructures processes, defines responsibilities, and anchors AI in data and decisions that matter.
The portrait by clusters reinforces this contrast:
On one hand, structured companies with formal strategy, dedicated responsibles, living ethical guidelines, ongoing training, and technical integration;
On the other, organizations that orbit isolated initiatives without institutional backing.
Among the large ones, the structured group is growing, but still below what is needed to compete in global ecosystems of models and data.
In parallel, the report reminds us that Brazil remains vulnerable if it limits its ambition to consuming external solutions. Technological dependence, when combined with poorly governed data, creates asymmetries that are difficult to reverse.
On the international stage, competitive advantage is emerging less and less from the brilliance of the model and increasingly from the strength of governance:
The European Union is advancing with a risk-based regime, with tiered obligations and expectations of transparency and security throughout the chain;
The USA combines executive directives, risk management profiles, and requirements for use case inventories in the public sector, driving agencies to professionalize responsibilities;
China adopts an approach that emphasizes labeling, synthetic content control, and provider accountability.
This is not about copying solutions, but about understanding that without a minimum framework, the conversation about AI detaches from reality. In Brazil, even though there is no specific AI law, the LGPD already operates as an immediate foundation for guiding data decisions, explaining criteria, and establishing responsibilities. It is a ready-made door that needs to be walked through with AI policies and practices.
The risks are concrete and do not lie solely in the reputational field. Poorly governed data leads to leaks, misuse, and regulatory restrictions that paralyze entire initiatives. Reputation suffers when systems hallucinate, discriminate, or fail to justify decisions affecting credit, health, benefits, or access to public services.
In the workplace, the promised productivity arrives, but asymmetrically: professionals who integrate AI into their daily work advance; those who wait for magical automation stagnate. Without intentional and continuous reskilling, asymmetries increase.
Given this, the report suggests a practical shift: less diffuse curiosity, more disciplined execution. A company that chooses real problems, such as well-defined pain points, available data, and clear metrics, moves from demonstration to results.
The one that establishes responsible leadership, creates a risk committee to arbitrate priorities, and publishes a usage policy sets the boundaries of the road;
The one that inventories data and use cases, classifies them by value and criticality, and plans technical integration with observability and model logging, repeating successes and quickly eliminating errors;
The one that trains leaders to make trade-offs and teams to specify, evaluate, and integrate, reduces dependence on heroes and diminishes Shadow AI;
The one that measures impact in time, cost, quality, and revenue, replaces vanity metrics with indicators that matter.
The implications spread throughout society. For citizens, automated decisions increasingly influence daily life, often invisibly. It is imperative to know one's rights, demand human review when affected by automated decisions, and cultivate verification habits in environments saturated with synthetic content. For professionals, the bar is raised: it is not enough to know tools; one must demonstrate applications with measurable results, master output evaluation, and integrate AI into the real workflow. Credentials that count are no longer just certificates but include portfolios with before-and-after, supported by metrics.
In companies, the discussion moves from the lab to the boardroom. The competition shifts to governance, integration, and ROI. Those who appoint responsible parties, define policies, control informal use, and anchor AI in reliable data create an environment where experiments become routines, and routines become advantages. Sector by sector, there are already mature paths:
Finance can reduce fraud and accelerate regulated services with traceability;
Retail can offer more accurate searches and recommendations without sacrificing human review;
Manufacturing can increase asset availability with assisted maintenance;
Healthcare can gain efficiency with clinical summarization in protected environments;
Governments can improve access to services, with transparency proportional to risk.
For the State, the message is strategic. Without a data policy and governance mechanisms, the country resigns itself to the role of a passive consumer of foreign infrastructures and models. With a high-value data policy (in health, education, climate, and mobility), robust anonymization, and clear access terms, public authorities can anchor research and applications of collective interest.
Government procurement that requires logs, robustness tests, and contingency plans encourages quality and creates predictability. Training programs for civil servants and educators, and regulatory sandboxes with metrics and deadlines, make the learning curve tangible. Labeling synthetic content in official communications signals standards and educates society.
Finally, there is the issue of practical sovereignty. This is not about reinventing language models from scratch but about building domain expertise in the layers where the marginal value lies for each context: well-governed sectoral data, applications embedded in critical processes, integration with legacy systems, and impact measurement. In other words, transforming AI from an object of fascination into an organic part of the country's productive and institutional infrastructure.
If we take the TEC.Institute + MIT report as a compass, the diagnosis converges: the bottleneck is not access to technology, but the institutional capacity to decide, govern, and integrate. Those who organize power, responsibility, and risk around AI will capture value sustainably. The rest will accumulate pilots, promises, and frustrations.
The near future will not be decided by the raw power of algorithms, but by the quality of the decisions we place between them and reality: clear rules, governed data, integrated processes, and capable people.
The window is open; closing it through indecision is the only choice that, over time, always costs more.
SQUADRA can help your company transform its legacy systems into strategic assets with artificial intelligence. Want to know how and why? Discover 7 benefits of unlocking innovation in your company and understand how SQUADRA does it.
