Episode transcript The transcript is generated automatically by Podscribe, Sonix, Otter and other electronic transcription services.
Hello everyone, here is Ricardo Vargas, and this is the 5 Minutes Podcast.
This episode came from a suggestion by Antonio Malagutti, a colleague I know from the PMI São Paulo Chapter. He told me that in a WhatsApp group on project management that has always been very active, the sharing of tips, courses and job openings has cooled down a bit this year. And he raised an idea that stayed with me: more and more, we are leaving human relationships aside to talk to artificial intelligence.
Think about how this happens in the daily life of a project. A question comes up about a contract, an estimate or a difficult meeting with the sponsor. In the past, many people would send a message to a more experienced colleague, or post the question in the group. Today, many people ask the AI first. The answer arrives in seconds, it is polite, it does not judge, and it does not ask for anything in return.
And that is exactly where the problem I want to discuss today begins.
There is one case that shows this with numbers. Stack Overflow is one of the largest question-and-answer websites for programmers in the world. For years, it was where developers from all over the world shared questions and solutions for free. A study published in 2024 in the scientific journal PNAS Nexus measured what happened to the site after ChatGPT was launched. In just six months, activity on Stack Overflow dropped by 25%, compared with similar platforms where access to ChatGPT was restricted. The decline had started some years before, but it became much stronger after generative AI arrived. By May 2025, the number of questions per month was back to the level of 2009, the year the site was launched.
And just to be clear, this episode is not sponsored by any of the companies I mention here.
AI models were trained, among other sources, on this open knowledge that people shared over many years. The authors of the study warned that knowledge is moving out of public spaces and into private conversations with AI. And in the future, there may not be enough public data to train the next models. In other words, AI learned from the people who shared. If we stop sharing, who will it learn from?
The same thing happens inside our organizations, except that nobody measures it.
AI answers generic questions very well, like how to build a risk matrix or write a project charter. But that is not what saves a project in a difficult week. What saves it is context. It is knowing that this client always approves changes at the last minute. It is knowing that this supplier failed at exactly the same point two years ago. It is knowing how a team in another area solved a similar problem, and what went wrong along the way. This knowledge lives in people's heads, and it only moves when someone asks and someone answers. When each of us talks only to our own AI, the team becomes a group of people sitting side by side, each one in a private conversation. Lessons stop flowing. And almost nobody notices, because each person, individually, keeps getting answers. Adam Grant, a professor at Wharton, helps us understand what is at stake in his 2013 book, Give and Take. He divides people into three groups. Givers, who help without keeping score. Takers, who try to get more than they give. And matchers, who give in the same measure they receive. In a study with salespeople, he found something curious: the people who helped others the most were both among the best performers and among the worst. What separated the two groups was boundaries. Those who reached the top helped a lot, but without letting others exploit them and without burning out.
Every community, and every project team, lives on this logic of giving to receive.
It starts to die when there are too many takers and not enough givers. And here is the provocation of this episode. AI has made it very comfortable to be a taker.
It never complains that we only ask, and it never asks what we brought to the conversation. For that very reason, it does not train us in the habit of exchange.
None of these means we should stop using AI, which is an extraordinary tool for our projects. But AI answers and does not remember us. Nobody recommends someone for a job because that person asked great questions to a chatbot.
Nobody stays late on a Friday to help save the project of someone who never answered a single message in the group. The trust that holds a project together in a crisis is built in these small exchanges, one at a time.
So I leave you with a question for this week. How many times did we ask AI something in the last seven days? And how many times did we answer someone else's question? If that balance is uneven, it may be worth answering a question in the group, sharing a lesson we learned, including what went wrong, or simply asking a colleague before asking the machine.
I hope you enjoyed this episode and see you next week with another 5 Minutes Podcast.
