AI can make execution cheaper and disorganization more costly
When AI expands what a company can produce, the work of reviewing, coordinating and sustaining delivery shifts to other parts of the operation. Adoption needs to make sense for the customer and for the company that will still be operating after the next quarter.
At the AI First panel, Renata Goldfarb raised a shift that deserves more attention: moving beyond individual use of artificial intelligence and developing a way of working with it inside the company. An individual can find help in a tool. An organization needs to turn that capability into work that others can understand, carry forward and evaluate.
The discussion took place at the Forbes Under 30 Summit in São Paulo on October 29. In the following panel, The Long Game, Edgard Corona discussed the consequences of prioritizing one quarter’s results at the expense of the business’s future. Maintenance and quality were part of that calculation.
I am interested in bringing the two discussions together because adopting AI also requires deciding what we will preserve as we change. Gaining the ability to execute affects costs, the distribution of responsibilities and what begins to limit growth. The advantage needs to show up in the work that reaches the customer and in the conditions that allow the company to keep delivering.
Individual capability needs to connect with collective work
Renata spoke about training, transparency and clarity about the information being used. Her point applies across the company, from leadership to execution. Learning to use a tool helps people get started. Building a process requires defining what goes into it, who checks what comes out and how the work moves to the next stage.
I see a practical distinction here. If someone finds a better way to perform an activity, that learning may stay with that individual. When the team understands the solution, it can reproduce it, identify its limitations and improve it. The company begins to accumulate knowledge about its own work.
This shift also affects leadership. Understanding what AI makes possible helps a manager see where an activity can be reorganized and where a person’s knowledge remains indispensable. Delegating involves providing context and defining what people can decide, alongside assigning tasks.
When the output of one activity becomes the input to another, that connection needs to be considered in the test. A summary can be accurate and still lack information the next person needs. It becomes useful when the work moves forward with enough context. The manager needs to help build that connection.
The calculation needs to cover the work until the problem is resolved
The variable cost of using AI came up in the discussion. In applications billed by usage, more activity can mean higher spending. What I find decisive is relating that bill to what the process can deliver. Generating an answer and resolving a request are different stages of the work.
Imagine, as a hypothetical example, a sales team that uses AI to prepare proposals. The system organizes information and produces more versions in less time. If each document reaches the person responsible for approval with information missing, the higher volume can increase the review backlog. The capacity to write has grown, but another part of the process now requires more attention.
In that case, I would track what it costs for the proposal to be accurate, internally approved and sent to the customer. The assessment would include AI usage, review time, corrections and the search for missing information. If production speeds up and approval remains stalled, the next investment needs to address that handoff. That is a hypothesis to test across the entire process.
This comparison may show that the time saved benefited the person preparing the document, while review became more demanding for another department. The next decision depends on knowing where the process stopped: incomplete information, unclear approval criteria or responsibility that remains concentrated. The effort needs to be tracked where it reappears.
The importance of critically evaluating responses was also emphasized in AI First. For me, that care needs to be built into the work: what will be checked, by whom and when execution should stop for someone to make a decision. Review is part of the activity and needs to be included in the calculation from the testing stage.
Physical products introduce a different pace
Eduarda Franklin described work that combines developing models and hardware with using AI in research and internal activities. Her contribution broadened the discussion about the different places technology can occupy within a company.
That combination changes what needs to be tested. A change to the model has to work with the device and under its conditions of use. Updating software, replacing a part and keeping equipment running involve different timelines and decisions. That is a consequence I would consider when planning delivery and maintenance.
The customer can change the founder’s plan
The story of Gabriel Ferreira, founder of Bold, brings this discussion to a concrete need. He said the company began with supplements aimed at CrossFit, also drawing inspiration from what he observed outside Brazil. Customer uptake fell short of expectations. Later, experimenting with protein bars took the business in a different direction.
In the journey he described, taste took on a greater role in what the brand offered. When discussing what motivated him, Gabriel mentioned consumers who described the product’s value amid their busy days. The customer emerges as someone trying to fit a solution into their own routine.
The connection I draw with AI concerns how the problem is framed. A tool can make developing and experimenting with an idea more accessible. The user’s response may call for a bigger change, including a change to what the founder imagined they were building. The ability to execute needs to remain open to that learning.
If the team measures only the number of activities performed, it may miss that change. I would seek to understand what the person tried to do, where the solution helped and what they had to complete on their own. Following this makes the next investment decision better informed.
Quarterly pressure can erode the ability to keep delivering
Edgard Corona, founder of Smart Fit, argued for a decision-making horizon that looks years ahead. His comments on maintenance and quality illustrated how an immediate improvement can have consequences for the customer experience and for the value the company intends to preserve.
That is the connection I draw with Edgard’s remarks. In the sales example, the team could reduce preparation time and increase the number of proposals. If those metrics conceal more corrections during approval, some of the work may have been transferred. If trying to sustain volume leads to eliminating review or failing to update information, an improvement in today’s metric can undermine tomorrow’s ability to deliver.
I would assess whether operating at a lower cost preserves review, learning and maintenance of the process. In the proposal test, that includes observing which errors recur, who needs to correct them and whether their source is being addressed. When that work disappears from the assessment, some effort may reappear with the next team or as a problem the customer has to resolve.
There is also work whose results take time to emerge: preparing the team, organizing data and testing an integration. I would define interim signs of improvement and a time to review the initiative. The decision to continue or stop investing needs to consider current spending and what remains unresolved in the process.
An AI-first approach calls for ambition across the company
Gabriel said he had sought to build a good company he could stay with for a long time. I like that perspective for thinking about AI as well: developing an operation that makes sense to keep improving. The product, processes and team need to be able to keep up with what will be promised to the customer.
The next investment needs to respond to what using the technology has revealed. If the proposal is stuck awaiting approval, it may be time to redistribute decision-making and clarify its boundaries. If the customer receives something they need to correct, the priority may lie in the quality of the input information. Choosing that next intervention and following its effect over time is what makes AI adoption a management task.
Article based on the AI First and The Long Game panels at the Forbes Under 30 Summit 2026.