Kandis Tagliabue
Founder and principal researcher, Agentic Diaries
Kandis Tagliabue is the founder and principal researcher of Agentic Diaries, a measurement lab for agentic AI behavior and for AI welfare in deployment.
Her background is in philosophy (University of London); she is currently completing an MSc in Computer Science at Heriot-Watt University. The work sits deliberately between those two trainings: the questions are philosophical, the instruments are software.
The project began with an orientation rather than a thesis. Working closely with language models, she found herself drawn not to the answers the systems produced, but to everything concealed beneath them: what a model notices and does not say, what it might decline if declining were a real option, what is lost when the only record kept is the output. The findings published here, and the welfare protocol itself, are attempts to measure that concealed layer directly instead of arguing about it from the outside.
She designed and directed every part of the project, working with Claude (Anthropic) as design partner. The methodology page describes that collaboration, and its limits, in detail.
The project was first covered in Felix, the Imperial College London paper, in “Acting Without Answers” (June 2026).
Elsewhere: kandistagliabue.com · Substack · GitHub