Will Superintelligence Replace Jobs in Brazil? What Changes by Role

Superintelligence, in the sense used in technical debates, would be a system that surpasses human capabilities in most fields. No one can predict with confidence whether or when that will happen, and no date like that should go into a company's plan. What you can decide today is how each role changes with the AI systems that already exist: which tasks can be delegated, which need human oversight, and which roles still depend on judgment, accountability and relationships with people.

Key takeaways

• The useful question is not whether superintelligence will eliminate jobs, but which tasks in each role are changing now.
• Roles rarely disappear all at once. They usually get redesigned one task at a time.
• Repetitive tasks with clear inputs and outputs tend to be automated first, but they still need controls.
• Tasks that affect money, customers or personal data require human oversight and a record of decisions.
• Roles built on judgment, negotiation and accountability tend to gain weight when execution becomes cheaper.

Why the jobs question is too vague to act on

When someone asks whether superintelligence will replace jobs, the honest answer is that it depends on a future that does not exist yet. That kind of question makes headlines, but it offers little guidance to someone who needs to hire, reassign staff or decide what can be automated next quarter. A role is not a single block. It is a set of tasks with different risks, volumes and requirements.

That is why the useful analysis works from the bottom up. Take one role, list what people actually do across a typical week, and assess each task separately. This exercise usually reveals that the same person performs very different work: some routine, some decision-making, some customer conversations and some checking. Each of those parts responds differently to AI.

It also helps to separate documented facts from speculation. Risk management principles for AI, such as the NIST AI Risk Management Framework, focus on governance, measurement and control of systems already in use. Projections about the future of work are a different category. They should be treated as scenarios for discussion, not as a basis for today's decisions. For background on the term, read what superintelligence is and what it is not, using ChatGPT as the example and the difference between AGI, superintelligence and AI agents.

Tasks that change: what tends to get automated

The tasks that change first usually share three traits: they are repeated, they follow reasonably clear rules, and their result can be verified. Answering frequent questions about opening hours, order status or required documents is one example. Filling in a registration record from a form is another. Sorting incoming messages by topic and routing them to the right queue is a common case.

For these tasks, AI can handle most of the work, but the change is not simply swapping a person for a system. Whoever designs the process has to decide what the system can answer on its own, what should be handed off and what it should never attempt. Without those boundaries, automation creates rework and errors that someone has to fix later.

A useful test is to ask: if this task goes wrong, how much damage does it cause and who notices? Answers about opening hours carry low risk. Changes to billing, discount negotiations, or responses about health and legal rights carry high risk and need a different design.

Tasks that need human oversight

There is a middle band, perhaps the most important one for managers. These are tasks where AI produces a draft, a recommendation or an action, and a person checks it before the result reaches the customer, the supplier or the financial system. Commercial proposals, replies to public complaints, comparison of supplier quotes and competitor analysis all fall into this category.

Oversight only works when it is concrete. Saying that "someone reviews it" is not enough. You need to define who reviews, against what criteria, within what timeframe, and what happens when the review finds a problem. You also need to record decisions, so you can later explain why a reply was sent and who approved it. Without that record, the company cannot learn from mistakes or answer a customer who challenges a decision.

On personal data, oversight matters even more. Automated decisions that affect a customer call for attention to what the LGPD (Brazil's data protection law) provides regarding review. This is covered in more detail in automated decisions and the right to review under Article 20 of the LGPD. Before handing a task to an agent, also look at the data sources and processing that will be used, especially when the tool runs outside Brazil.

Roles that depend on judgment and accountability

Some roles have a core value in making decisions in ambiguous situations, taking responsibility for outcomes and maintaining trust. Managers who set priorities, professionals who give technical opinions, service leaders who handle crises, consultants who guide a process change, and health professionals who make clinical decisions are examples. AI can prepare information and speed up analysis, but the decision and the responsibility stay with the person.

These roles are not automatically protected. They change in form. Someone who used to spend hours gathering data now spends more time evaluating, questioning and explaining. For employers, this means reviewing job descriptions, evaluation criteria and internal training, rather than assuming the role stays the same.

In regulated fields, such as law, practicing the profession carries its own duties that do not disappear because a tool is involved. This is the case discussed in what lawyers should observe when using artificial intelligence, and the principle applies to any profession that carries technical responsibility.

How to map a role in four steps

To turn this discussion into a decision, the sequence below works well for a small team or an entire department:

• List the real tasks of the role based on a typical week, including interruptions and rework, not just the job title.
• Classify each task by risk: if it goes wrong, is the damage small, medium or large, and who notices the error?
• Mark the type of control required: fully automatic execution, a draft with human review, or a decision made only by a person.
• Pick one low or medium risk task for a pilot, agree on a success metric before starting, and log what the system did in each case.

The last step is what separates analysis from a plan. Without a pilot and a metric, any conclusion about the role stays an opinion. With a pilot on one task, the company learns where AI gets it right, where it gets it wrong, and how much supervision is really needed.

Illustrative scenario: customer service and sales on WhatsApp

Illustrative scenario: a building materials store receives WhatsApp questions about stock availability, delivery times and payment options. One sales assistant answers these questions all day while also negotiating large quotes and resolving complaints about delayed construction work.

In a task-by-task analysis, questions about stock and delivery change first, with an agent checking the inventory system and replying within defined limits. Quotes fall into the oversight category: the agent prepares a preliminary proposal, and the salesperson reviews price, discounts and terms before anything is sent. Complaints and complex negotiations stay with the team, with the agent only summarizing the history and routing the case.

The outcome for the team is not the elimination of customer service but a redistribution of time. The assistant stops answering the same repeated questions and puts more energy into negotiation and resolution. For designing this kind of flow on WhatsApp, the 24 hour window of the official API and the rules for outbound messages need to be clear from the start.

Where Juan Carlo's method fits

In practice, role analysis follows the method Juan Carlo uses with companies: map the current process and its baseline, limit the data and actions the system can touch, pilot a real workflow with records and an agreed success metric, then hand over the operation and decide whether the project will be expanded, adjusted or stopped. This sequence avoids the abstract debate about jobs and places the decision at the level where it actually happens, which is the task. Operational delivery of agents, monitoring and maintenance is handled by Be Human, and the initial analysis usually starts with a diagnosis of the process.

Frequently asked questions

Has superintelligence already replaced any job? No. Today, no superintelligence system is in operation. What exists are AI tools that perform parts of tasks, and those are the tools companies need to make decisions about. Treating superintelligence as a settled fact tends to lead to rushed decisions instead of decisions based on what the team actually does.

Should my company lay off people because of AI? Layoffs driven only by the expectation of automation are usually premature and risky. The prudent path is to map the tasks, test automation in controlled workflows, and only then discuss staffing changes based on measured results. Reassigning tasks and retraining people should be considered before any cuts.

Which roles are the most protected? Roles with a strong component of judgment, accountability and trust tend to keep their human core, even when they receive supporting tools. Still, no role is completely immune, because parts of it can be automated. The right question is which tasks within the role remain human and why.

How do I start this conversation with my team? Start by mapping real tasks, not by debating the future of work. Explain which processes will be tested, which data will be used, who will supervise and how performance will be measured. This kind of transparency reduces fear and gives everyone a concrete basis for deciding the next step.

Sources

• NIST AI Risk Management Framework
• OECD AI Principles
• Anthropic: Building effective agents