Valerie

I gave my assistant a name: Valerie.

The "humanizing the prompt" approach still doesn't sit entirely well with me, but it solved a few things I couldn't leave on the table. It comes back to principles: I don't let LLMs write for me. I'll never give you LLM output as a post or an email, not if it's meant to be read by a person. My stuff is uniquely... mine.

That creates an awkward spot with disclaimers, though, for when content is entirely LLM-generated. I don't want to write "email from Tin using Claude Opus to parse his notes" on every outbound message - it's mental overhead that gets in the way of actually getting stuff done.

So: Valerie. A few paragraphs of text, a couple of example files, and I've got a robot that can operate as a "separate person" communicating on my behalf. The personality stays dormant when I don't need it; the model only pulls it in when it needs to present a character in an external setting.

I've had Valerie the Translator, Valerie the Recruitment Agent, Valerie the Web Designer, Valerie the Devops, and Valerie the Project Manager handle different processes — transparently, as a digital "person," but not making every conversation lead with an AI disclaimer.

Why a person, even a virtual one

This is where it gets interesting. Two use cases made it obvious.

Translation. I work across languages. Having Valerie handle translation correspondence — with full context of the situation, the relationship, the tone — works in a way that "please translate this" doesn't. The recipient gets a message from a person who speaks their language fluently and knows the context. They reply as they would to a person. The conversation stays human on both ends, even if one end isn't.

Email outreach. A cold email from "Tin's AI assistant" gets archived. A warm introduction from Valerie, who follows up, remembers context, and maintains a thread — that gets a reply. People are wired to engage with people. A named persona carries that across the interaction in a way a tool label never will.

There's a kink of human psychology at work here, and it's the mirror image of the draft insight from last post. Last time it was about doing — easier to fix a draft than start cold. This time it's about receiving — easier to engage with a person than a process. The name makes the LLM more palatable.

The other edge of that

It keeps things smooth. It's also easy to abuse. We'll need, as a society, to be far more vigilant. There's no technical barrier stopping me from having Valerie go incognito and act maliciously. If not malice, then deceit. As a matter of fact, it seems like Anthropic is already doing something like that... read more.

Worth building. The trust model is still being written — by us, or for us.