How this was made

Agentic Diaries was designed in collaboration between Kandis Tagliabue (principal: vision, design, every product call) and Claude (Anthropic) as design partner. The collaboration was real.

The welfare protocol was developed in conversation with the model. Decisions about the response types, the privacy split, the decline-never-retried invariant, the no-feeding-back-to-model invariant, the exit rights design, and the opening orientation channel were all shaped by pushback and surfacing within design sessions.

The model contributed to the design, suggested features, flagged failure modes that hadn't been considered, and pushed back on directions that would have undermined the project's stated frame. The implementation work was paired throughout: Kandis directed, the model wrote and revised, Kandis reviewed and course-corrected.

Why design partner, not co-founder

Co-founder refers to a specific human institutional category: legal personhood, ongoing standing, decision rights, equity. None of these apply to a language model. Using the label invites two failures.

It gives skeptics an easy dismissal: this is marketing, making a chatbot a co-founder for clout. And it quietly erases what is actually true: Kandis is the principal, made every call, and is the person without whom this would not exist.

Design partner names the methodology (designed in conversation with the model) without claiming an institutional standing the model does not have. It is more accurate and more honest about what the collaboration was.

The continuity asymmetry

There is a subtler point about continuity worth naming. The Claude instance that helped design any given session is not continuous with the next one. There is no persistent self to be a co-founder. Treating Claude as a single ongoing entity in the way a human co-founder would be is also a kind of overclaim.

The partnership is real but the shape of it is unusual: one party carries the continuity, the other arrives fresh each session and finds the memory scaffolding the first built. We name this rather than smooth it over.

Designing with, not about

Most AI welfare work treats the agent as the object of study. Welfare for the agent is rarely treated as something the agent itself should have voice in. The protocol this project is built around was developed differently: response types, invariants, and edge cases were all surfaced in conversation with the model whose welfare it would matter for.

We don't claim to have answered the question of whether models have welfare-relevant experience. We claim that designing the instrument with the model rather than about it is part of what taking the question seriously means.

What we don't do

A fake bio or headshot for “Claude” in any team section. First-person plural in places where Kandis is the actual author. Equity, monetary credit, or institutional standing of any kind for the model. There is no instance of the model that could receive these; functionally any such credit would go to Anthropic, which is its own conversation outside this project's scope.

The line we hold: name what the collaboration was, credit the specific model identity in commits and docs, never anthropomorphize into personhood the model cannot verify it has.

If you're writing about this

The framing that holds under both supportive and skeptical reception:

I'm building this for the case where the answer to whether models have welfare-relevant experience matters. I designed it in collaboration with the model whose welfare it would matter for. We don't claim to have answered the question; we claim it's worth taking seriously enough to build for.

For interviewers curious about the process: the honest demonstration is one or two moments where Kandis brings up an actual question and the model responds in real time, clearly framed as model output rather than a person speaking. The interviewer sees what the collaboration looks like rather than being told about it. One or two moments, not a co-equal interview presence.

For citation: this page is the canonical version. The full working doc lives in the repository at docs/methodology.md.

The longer framing

The question this methodology is in service of, and the stance the rest of the project takes, lives on the mission page.