ABOUT THIS EPISODE
I built an AI team to help me make a podcast. Then I heard the episode and wanted to start again.
The research was interesting. The script had been fact-checked. There was even a model of my voice reading it. But it didn't feel like me. Was AI helping me find the story—or helping me keep improving a version I didn't actually want?
For the first episode of Growth Sold Separately, I follow that question through classrooms, an MIT essay-writing study, and experiments in creative work. Better results with a tool, what we can do without it, and whether we like what we've made turn out to be different questions. Making this episode put me right in the middle of them.
I'm Tony Lee. This is a story about the work I wanted to hand off, the decisions I wanted to keep, and the version I finally wanted you to hear.
CHAPTERS
00:00 The monster is pointing at me
02:09 Close the laptop
05:38 What MIT measured
10:01 Make it sound like me
13:03 Another version
16:24 Good by whose measure?
18:28 The missing Tony
21:58 What I want to keep
PRODUCTION
This episode combines my own recordings with AI-generated narration using a model of my voice. I developed the script with AI, reviewed the research and approved the finished episode. Instrumental music was generated with ElevenLabs.
SOURCES AND FURTHER READING
The studies below examine particular tasks and settings. They do not establish whether making this podcast made me smarter or dumber.
Bastani et al. (2025), Generative AI without guardrails can harm learning. Mathematics practice and the exam without AI:
Source
Kosmyna et al. (2025), Your Brain on ChatGPT. Essay-writing, EEG and recall; a preprint, not an intelligence test:
Source
Kestin et al. (2025), AI tutoring outperforms in-class active learning. Physics tutoring and immediate assessments:
Source
Thoeni and Fryer (2026), AI Chatbots in Higher Education. Why twelve weeks of availability is not twelve weeks of use:
Source
Hou et al. (2025), The Double-Edged Roles of Generative AI in the Creative Process. Help at different stages of design:
Source
Wadinambiarachchi et al. (2024), The Effects of Generative AI on Design Fixation and Divergent Thinking. The chatbot-avatar experiment:
Source
Doshi and Hauser (2024), Generative AI enhances individual creativity but reduces the collective diversity of novel content. Creative ratings and similarity in short fiction:
Source
Baird et al. (2012), Inspired by Distraction. Returning to previously encountered creative problems after a break:
Source
Later counterevidence: Murray et al. challenged the proposed mind-wandering benefit. The explanation remains unsettled:
Source
Coverage discussed in the episode:
Rael Hornby, Laptop Mag (July 2025):
Source
Erik Ofgang, Tech & Learning (July 2025):
Source
Questions or a source I should examine? Leave a comment on the episode.
