Zendesk chief executive Tom Eggemeier has said an experimental AI project with his son exposed the limits of “vibe coding” and changed the way he thinks about developing products and training staff.
Eggemeier worked with his son Will, a Boston College senior and basketball player studying history with a business minor, to create Underdog AI. The predictive analytics app is intended to help college athletics programmes identify players in the NCAA transfer portal before wealthier schools recruit them.
Will proposed the idea earlier this year after his father had repeatedly told his children that AI would be “bigger than the printing press, the internet, the Industrial Revolution”. The pair wanted to build a tool that could help “level that playing field a little bit”, Will said.
The early stages of development moved quickly. But Will became concerned when the system began making recruitment recommendations that did not appear credible. The speed of development also created technical debt involving security and API protocols, leading to hours of debugging.
Eggemeier said the experience demonstrated that there were no shortcuts to building software that is both seamless and secure. Two college basketball general managers told the pair that the application matched about 70% of products already on the market.
Will cannot sell the tool to rival schools while he remains on Boston College’s roster, so he is continuing to refine it until he graduates. He hopes to pursue a career in the sports business and believes that being able to deploy AI will be essential to his success.
Lessons from the AI project
Eggemeier said working alongside his son had also influenced his leadership at Zendesk, making him “a better leader for our R&D organization, our product, and AI”. He said the project had shifted his focus towards how AI agents could reduce costs, save time, cut complexity and solve practical problems.
He urged other business leaders to use the technology themselves. Eggemeier said he had used a personal agent to renegotiate his insurance, internet and mobile bills, as well as his subscriptions, saving $3,150 a year.
He also trained ChatGPT Work to write in his voice, but noticed it repeatedly ignored his instruction not to use contractions. Eggemeier said he believed contractions could be harder for non-native English speakers to understand, and that the repeated error had prompted him to think more deeply about how memory and skills affect products and the internal use of AI.
Eggemeier said companies should continue hiring young employees, whose familiarity with AI can create a productive dynamic alongside experienced workers. He also advised employers to acknowledge anxiety about the technology, help staff develop new skills and create projects focused on genuine problems.
For families, his advice was simpler: “And if your kids want to build something with you, do it.”
