Episode Details

After a frustrating week of trying to wrangle AI outputs, we decided to explore the risks of overreliance on AI. It’s good for us to question our tools. It enhances our processes and challenges us to find the right tools.

Listen on Spotify | Listen on Apple Podcasts | Watch on YouTube

In this episode, we say the quiet parts out loud. Not only are LLMs often feeding us incorrect information, but over-trusting these systems poses a serious risk.

We can look at this Rolling Stone article headline and immediately laugh it off. It is insane to believe this will happen to anyone we know. However, in Mark Zuckerberg’s vision of the AI future, your friends will be bots.

The loneliness epidemic is real. One in three Americans feels lonely every week. Data from Harvard’s Making Caring Common Project supports that loneliness is tied to increasing feelings of anxiety, not part of this country, and being about more than social isolation. 65% of respondents blame “our society,” pointing to a lack of confidence in our way of life and institutions.

So, it should be no surprise that Harvard Business Review found that the top three use cases of 2025 involved loneliness and navigating life's stresses.

AI could quickly become the next addiction for a world desperate for solutions. The fact that there’s demand for robo-companionship shouldn’t be treated as validation for building more tools to disassociate them from life.

Let’s go back to exploring this topic from the perspective of business users.

Understanding GenAI’s Productivity Gains

As we barrel into the AI-powered era, we can take one of two perspectives:

* GenAI products are the next evolution of SaaS: Precise tools for specific workflows

* LLMs are the next evolution of social media, where instead of degrading our interpersonal relationships, AI will addict us to easy and often incorrect information

The majority of the research identified productivity gains and time savings, which would support the goal of GenAI as a professional advantage.

But when you dig into the data, there are concerns.

Many are funded by Microsoft, OpenAI, and Google, like this one showing that GitHub Copilot users completed tasks 55.8% faster than the control group. While that result was impressive, they were being assessed on their ability to complete a very basic task. The paper’s results were boosted by showing that people with less experience benefit more from coding assistants, something that should worry anyone concerned about being replaced by cheaper talent who are boosted by AI.

But these results were refuted in a separate study where no DevOps productivity gains were found from usi