Building Claud Projects: How Specialized AI Changed the Way I Work

At some point last year, I started building characters with AI.

At first it was fictional chatbots A character from a book to chat with. But then i realized it actually could be useful in my work. So I created a few custom Claudes—each with a clear personality, purpose, and way of thinking.

The Translator

The first one was simple in theory: an English → Russian translator. But I didn’t want Google Translate energy, I wanted something that understood lyrical voice and subtext, adhere to literary rhythm.

So I shaped it carefully:

  • It preserved tone instead of flattening it

  • It knew when dialogue should sound sharp, tender, or restrained

  • It could switch between literal accuracy and literary intention

The quality of the output really surprised me. I didn’t feel like I was losing parts of myself when moving between languages. The translator held context and respected the emotional weight of what I was saying.

Dave, The Founder

Decisive, calm, slightly blunt—the kind of person who doesn’t spiral, but asks one sharp question and cuts straight through the noise. I didn’t explicitly “create” him at first; he emerged naturally through the way I prompted, challenged, and refined responses, until over time he became consistent enough to feel like a presence. When I was stuck, I’d ask Dave; when I was overthinking, he’d call it out; and when I needed to pressure-test an idea, he didn’t flatter me—he clarified me. What struck me most was that I wasn’t outsourcing my thinking, but externalizing a mode of thinking I already had, one I couldn’t always access under stress, doubt, or fatigue.

Design Critique Bot

I also built a design critique bot for myself—because giving and receiving feedback is a repetitive task, but still one that requires care. I wanted something that could consistently challenge my decisions, ask the uncomfortable questions, and point out weak spots without ego or fatigue. It doesn’t replace human critique, but it sharpens my thinking before I ever bring work to a room. In practice, it’s like having a calm, always-available design partner who never phones it in.

The Result

It’s kind of incredible to have a technology at your fingertips that:

  • doesn’t require constant re-prompting

  • already understands the task

  • and does exactly what you need it to do

I didn’t lose agency by creating these Claude projects—I gained clarity. We talk about AI as productivity software, but I think its most interesting future is more human than that: a second brain that doesn’t panic, a translator that doesn’t erase nuance, a founder voice that steadies you when confidence wobbles. Not because it knows more than you, but because it helps you hear yourself more clearly. And honestly, that might be the most powerful tool I’ve ever designed.

Next
Next

Claude Code Hack Day at Amazon at AWS Builder Loft