Artificial Intelligence
September 03, 2026
The current challenge of collaboration
As the creative process begins with AI, organizations need to rethink how to preserve the collective construction of ideas
In just over three years, generative artificial intelligence tools have come to occupy a large part of the work in organizations, whether for writing, programming, researching, planning, analyzing, or designing.
The presence is already so great that, for many professionals, it has become difficult to imagine what work would be like without access to these models. Chats, copilots, and agents have become so present in the corporate daily routine that they are already approaching the logic of expected applications in a basic “package” of administrative tools, like MS Office in the early 2000s.
Productivity gains help explain this adoption. They appear both in the almost immediate results we perceive after a few prompts in our routine and in formal studies. In a Microsoft Research study, developers with access to Copilot completed a programming task, on average, in less than half the time spent by the group without the tool.
When a technology so expressively shortens the distance between idea and execution, it doesn’t just improve efficiency. It also alters the expectation about delivery speed, expands the possible volume of work, and changes the perception of what is considered fast, sufficient, or competitive within organizations. Even our understanding of what a complex task is begins to change.
This logic is not limited to a specific area. As assistants become more capable and agents begin to execute tasks in an integrated manner, models cease to act solely as interactive interfaces and begin to compose, with greater autonomy, the work operation itself. Instead of merely suggesting paths, they also begin to execute parts of the delivery.
A group of researchers, with participation from Tsinghua University in China, already suggests a model in which the systems themselves organize part of the flow and only resort to people when they need judgment, intention, limits, or subjective decisions. We are not just talking about tools that help create, but about systems that are also beginning to organize and execute.
Interactions
But deliveries are not the only part of the job. Interactions are also fundamental. Jaime DeLanghe summarizes this idea well in a text called “Work is Conversation”: much of the work happens precisely in the exchanges, alignments, questions, disagreements, and decisions built between people. And this is where a less obvious change emerges. While systems do more, a large part of interactions with AI remains individual.
Currently, most AI model interfaces operate through individual interactions. The most common pattern is still a single person in front of a chat, writing prompts, receiving answers, refining paths, and advancing alone. Even when these tools are integrated into other platforms or involve multiple agents operating together, the central logic remains a direct relationship between individual and system. The experience continues to be, for the most part, a private conversation.
This interface works very well. It reduces friction, simplifies usage, and transforms complex tasks into accessible interactions, including by voice today. But this efficiency also shifts a significant part of creation to a less visible and less shared sphere. Ideas begin to be explored in private conversations with models.
Early versions emerge quickly. Structures, drafts, prototypes, and arguments take shape even before other people join the conversation. Boris Cherny, creator and leader of Claude Code at Anthropic, publicly describes a workflow where the tool participates from the planning of functionalities and is integrated into the company’s daily development.
The creative process that often began with interaction between people now begins with interaction between the individual and the system. Teams still exist, evidently, but in many cases they enter later, when a direction has already been chosen, a hypothesis has already been organized, and a solution has already taken shape.
It is precisely at this point that the discussion about collaboration becomes more delicate, because collaborating has never been just about dividing tasks. Collaborating is also a way of thinking beyond one’s own individual vision. It is in the exchange that fragile hypotheses find counterpoints, different repertoires intersect, blind spots appear, and decisions gain more density.
Collaboration also relates to learning. Malcolm Knowles, by popularizing andragogy, helped consolidate the idea that adults learn best when they can relate experience, autonomy, and meaningful contexts of exchange. Albert Bandura, in turn, showed how much of human behavior is learned through observation and interaction with other people.
In simple terms, we learn a lot from other people. When a relevant part of the initial exploration stops occurring in a group, it’s not just a conversation that is lost. Context, shared repertoire, and the opportunity to follow how an idea was built are also lost.
Fewer collective constructions
The risk is losing collective constructions. When fewer people participate in the beginning of a process, less repertoire circulates, less context is shared, and less space exists for an idea to be tested, expanded, or revised before gaining momentum. Execution may become faster, but this can happen at the expense of the quality of the thinking that sustains that execution. And, along with this, part of the thinking systemic is also lost.
Luiz Carlos da Vila, a great Brazilian samba artist, used to say humorously that “man’s greatest invention is the wheel.” He wasn’t just referring to the object, but also to the circle of conversation and the samba circle, genuine spaces for interaction, creativity, and exchange. The image helps us remember that not all work gains appear in a delivery. There is also value in the encounter, in the circulation of ideas, and in what emerges between people.
This point gains even more importance at a time of anxiety regarding technology. The Workmonitor 2026, by Randstad, shows workers trying to adapt to the growth of AI while also dealing with concerns about its effects on work. Under pressure to keep up with this change, producing first and discussing later may seem like the safest path. But speed alone does not guarantee value.
The rush to execute, mediated by models capable of generating texts, media, or analyses in seconds, can favor what is already being called AI slop or, in the workplace, workslop. In simple terms, these are AI-generated deliverables that appear professional at first glance but are superficial, inaccurate, or incomplete and end up requiring human rework. An individual’s productivity gain can thus transfer work to others. What seemed like individual efficiency turns into a collective cost.
When creation happens individually with models, the risk of sycophancy also arises. This is a behavior in which systems favor responses that seem to please the user or validate their initial hypotheses. Research indicates that, in these situations, AIs often reinforce ideas and commands instead of questioning them. The problem is that, instead of putting an idea to the test, they often just process it and return it with an appearance of consistency. AI can amplify our ideas, but it doesn’t always confront them.
Practices that attempt to compensate for this behavior are already beginning to appear. One example is commands or skills like “Grill Me”, which ask the model to interrogate the user, question premises, and avoid immediately proceeding with the first solution. This is useful because it returns part of the decisions to the human and creates more space for reflection. Even so, it does not completely solve the problem of collaboration: a more critical interaction between person and model remains an interaction between a person and a model.
This seems central to thinking about the current moment. I am a clear advocate for the use of this technology. I have even written about the importance of preparing the organizational environment, organizing data, and expanding literacy so that the use of AI is more conscious. I revisit these lenses here because the issue of collaboration passes through the same point.
One of the challenges of management today is balancing speed with safety and quality with autonomy. It is not enough to adopt the tool. It is necessary to understand what it accelerates, what it hides, and what human capabilities may be being displaced from the process without the organization immediately realizing it.
Much is said about keeping the human in the loop. Something I support, but perhaps it is no longer enough to describe the size of the challenge. The point now is not just to keep a human interacting with increasingly capable systems. The point is to keep humans, in the plural, within the sense-making process.
This means ensuring that groups continue debating, disagreeing, revising premises, sharing context, and maturing decisions collectively. Preserving spaces for contrast, listening, and shared construction becomes an essential part of the quality of the work itself.
AI undoubtedly creates a real opportunity to reorganize work. Productivity gains show this. And perhaps this is precisely the most important point. If systems can execute more, we should use this gain to enhance the quality of human participation, not to further reduce the moments when people think together. With more time available for certain tasks, we should be able to consider more views, involve more perspectives, and build solutions with greater collective density.
But this reorganization will not happen on its own. It requires a more mature understanding that, in addition to individual decision-making, collective collaboration remains irreplaceable. In a scenario where interactions with AI are mostly individual and private, one of the most important tasks for leaders will be to recreate conditions so that collective thinking does not become a late, protocol-driven, or merely decorative stage.
Perhaps the most important question at this moment is not whether artificial intelligence will collaborate with us. In many cases, it already does. The more difficult question is another: amidst so much individual efficiency, will we continue creating conditions for humans to keep collaborating with each other?
| To access the references of this text click here |
Who wrote this column
Lucas Tangi
