Keeping up with the advances of technology requires deciding where to concentrate attention. A tool can gain new functions while the team is still learning to use the previous version. Between getting to know something new and incorporating it into the work, there is an effort of study, testing, and adaptation that the launch announcement does not show.

It is in that difference of rhythms that professional updating becomes difficult. The workday stays occupied, the processes need to work, and each change competes for time with responsibilities that already exist.

One way to deal with that pressure is to separate three decisions: what to follow, what to learn, and what to adopt. This article gathers research on artificial intelligence, professional skills, and adult education in order to discuss that choice. The practical guidance is a proposal for organization, with no promise of productivity or of complete command of the subject.

Technical progress has speed, but it also has limits

Artificial intelligence offers a documented example of rapid change. Stanford’s AI Index 2026 records an advance of 30 percentage points in one year on Humanity’s Last Exam, an evaluation with high-difficulty questions. The same chapter describes uneven capabilities: strong results in mathematics coexist with difficulties interpreting analog clocks. Source: Stanford HAI, technical performance in the AI Index 2026.

These results show progress on specific tasks. They do not measure the speed of all technology, nor do they guarantee that a new tool will improve any routine. A good result on an evaluation needs to be set against the problem in which the system will be used.

For someone at work, that distinction changes the question. Before investing hours in something new, it is worth identifying which task it can improve and how that improvement will be verified. The demonstration of a function sparks interest; the evaluation of use needs to include what happens when the data are incomplete or the answer is wrong.

Skills change without erasing all previous knowledge

The Future of Jobs Report 2025, from the World Economic Forum, gathers the expectations of more than a thousand employers. The estimate is that 39% of existing skills will be transformed or become outdated between 2025 and 2030. The report itself records a reduction relative to the 44% estimated in the 2023 edition. These are projections about different economic and technological changes, not a measurement of knowledge already lost because of AI. Source: conclusions of the 2025 report.

The reading requires care. A forecast of change in skills does not determine which knowledge each professional should abandon. That decision depends on the activity, the tools used, and the responsibilities taken on.

In an illustrative example, an analyst who receives a new reporting tool needs to learn its commands. They will still have to check the origin of the data, understand the calculation rule, and explain the result. Learning the interface is added to the knowledge required to judge the work produced.

For that reason, the study plan can combine two fronts: the concepts that sustain the activity and the new functions that help carry it out. Learning to verify a piece of information continues to make sense when the lookup tool changes.

Lack of time is also a barrier to access

The difficulty of learning cannot be reduced to individual willingness. In the Trends in Adult Learning report, published in 2025 with data from the 2023 skills survey, the OECD points to work and family responsibilities as the main barrier for 48% of the adults who faced obstacles to the training they wanted. The scope considers people from 25 to 65 years old and the average of the countries participating in the OECD. The percentage does not represent all adults, nor a specific estimate for Brazil. Source: OECD, barriers to adult learning.

That result helps in understanding the limit of a common recommendation: “just study more.” When the update is necessary for the work, the team’s planning needs to include learning time. Adding courses to an already full schedule leaves unanswered the question of when the study will happen.

A viable proposal should make explicit what will be learned, which activity will have reduced room, and who will be able to help with questions. It also needs to accept different rhythms of familiarity with the tool.

Following, learning, and adopting are different decisions

A news item can deserve attention without requiring a course. A course can be useful without justifying the immediate replacement of a system. Separating those decisions makes it possible to define the next step with more clarity.

The division below is a practical proposal for organizing what is new. It is not a method validated by the reports cited.

  • Follow: there is a relationship with the field, but a concrete application is still missing. The expected result is to record the source and the reason for returning to the subject.
  • Learn: there is a relevant task that requires a new skill. The objective is to be able to carry out an exercise and explain its limits.
  • Adopt: a test demonstrated usefulness and the conditions of use were evaluated. The next step is to define who is responsible, guidance for use, and a review criterion.

The criterion can begin with a simple sentence: “we need to learn this in order to solve this problem.” If the problem still cannot be described, following the topic may be enough at that moment.

It is also worth recording what was left for later. A short list, with a reason and a condition for taking it up again, makes it possible to reconsider the decision when a need appears. The arrival of another announcement, by itself, does not have to change the study plan.

A learning cycle that produces something verifiable

Study can begin with a small task, with an observable result. The proposal below serves as a starting point and should be adjusted to the time available and to the complexity of the work.

  1. Choose a recurring problem. Describe the current difficulty, such as checking fields in a spreadsheet or organizing information for a report. The scope should fit in one exercise.
  2. Consult one main source. Read the documentation or the training material tied to the function chosen. Note the date, the version when it is relevant, and what is still not clear.
  3. Practice with safe examples. Use fictional data and include situations that can provoke an error. Define beforehand what a correct answer would be.
  4. Check the complete result. Consider the effort of preparing the input, checking the output, and correcting problems. A fast answer that requires a long review may not meet the need.
  5. Record the decision. Write what worked, where there was difficulty, and whether it is worth continuing, adopting, or ending the experiment.

Imagine, as an illustrative example, a team evaluating AI for summarizing documents. The exercise can use a fictional document with a deliberate contradiction. Besides checking the quality of the summary, the team observes whether the contradiction was preserved, omitted, or presented as a resolved fact.

At the end, the learning should make it possible to explain under which conditions the tool helps and which check remains necessary. The test can also justify the decision to keep the current process. The article about what changes in free Gemini deals with which tasks to evaluate before paying for a subscription.

The team needs to reserve room to learn

An organization that requires updating can make that requirement concrete. Some measures to consider are reserving study time in the workday, limiting simultaneous changes, and designating a reference person for each tool adopted.

It is also fitting to agree on what will be evaluated. Instead of using only the number of courses completed, the team can observe whether the professional can carry out the proposed task, recognize an error, and ask for help when necessary. The choice of those criteria depends on the work; the reports cited do not establish a universal study-hour load.

The record of the learning deserves attention. At Fluxonar, the article on project memory for AI describes how decisions and lessons are kept in a file of the project. It is a concrete example of reusable documentation, without depending on memorizing every detail in a work session.

The idea can guide a simple team practice: document the problem, the solution adopted, and the reason for the choice. A list of installed tools says little about what was learned with them.

Updating can have a direction

Choosing priorities requires reviewing them. A tool that was once dispensable may come to solve an important problem; a familiar process may stop meeting the needs. Keeping review criteria prevents turning a provisional choice into a permanent habit.

The next step can be small: select a real difficulty, reserve a possible period of study, and produce a result that another person can check. From that, it becomes clearer what to go deeper on and what can wait.

The speed of the launches belongs to the market. The learning plan needs to fit the work and produce enough understanding to decide with responsibility.

Sources and scope of the analysis

Sources consulted on September 22, 2026:

This is an analysis of public sources, accompanied by proposals for organization. The examples of an analyst and a team are illustrative. The surveys have distinct populations, periods, and objectives; they do not allow calculating a single rate of technological acceleration or predicting the future of an individual profession.