The First AI-Native Generation.
A generation is coming of age inside a technology nobody has finished understanding. The Center is asking what sustained interaction with conversational AI does to cognition, identity and agency, and publishing what it finds.
What is already known
The survey layer is crowded and it is worth saying so. This study is not another sentiment poll.
Gallup, with the Walton Family Foundation and GSV Ventures, April 2026
n = 1,572, ages 14 to 29. Excitement about AI down 14 points. Anger at 31 percent.
Wharton, with Gallup
79 percent of Gen Z say AI causes laziness.
TalentLMS
47 percent of Gen Z employees say they get better guidance from AI than from their managers.
Deloitte runs an annual Gen Z survey. KPMG runs an intern pulse. Between them the question of how this generation feels about AI is well covered, and the answers are converging.
Three gaps, and they are the reason this study exists.
Nobody is doing qualitative depth.
Everything published is quantitative sentiment. There is very little work on how AI-native people actually reason, how they decide what to delegate, and how they structure work when a machine is always available.
The datasets are US-centered.
A Canadian qualitative cut does not exist.
Nobody has asked the researchers.
The people building and studying these systems have views about the generation growing up inside them, and those views have not been collected in one place.
Method, stated before the findings
Two layers of evidence, gathered differently and never merged.
What experts believe
Long-form interviews with AI and machine learning researchers through the Co-Existing with AI series, weighted toward the Amii, Mila and CIFAR network. Coded against a fixed frame and published as an elite-informant qualitative study.
What AI-natives actually do
New fieldwork, because researchers are not AI-natives and the corpus cannot answer for them. Open now.
The layers never merge
A study that blurs expert belief into lived behavior is describing neither. Keeping them apart is what makes either one readable.
Long-form interviews rather than survey instruments, coded against a fixed frame: what current systems cannot do, where the field is overconfident, what changes about human work and judgment, what the public misunderstands, and where informants disagree with each other. The disagreements are the finding.
Podcast and long-form interview material is treated as data under an established protocol, following Kulkov, Kulkova, Rohrbeck and Menvielle in the International Journal of Qualitative Methods (2024). Sample adequacy follows Guest, Bunce and Johnson in Field Methods (2006), which found thematic saturation within the first twelve interviews.
Two limitations, named here rather than buried. Informants were invited rather than sampled, so selection is not random. And these were conversations rather than protocol-driven interviews, so interviewer effect is present. Both will be stated in the published methods section.
What gets published
The report
Thirty to fifty pages on what the evidence supports, what it does not, and where the open questions are. Full quote appendix with sources.
The policy brief
Six to ten pages for the people who write rules, written so they can act on it.
The framework
A developmentally responsible AI framework organizations can apply without a research team.
Published free, as every comparable flagship is. First outputs November 2026.
Co-Existing with AI