Nice Why
A tiny studio doing

branding, design & digital
For people who actually care
2025

Aidan 2.0

I built an AI clone of myself. Weird, right?
Preset questions help save on API costs
Python
Django
Cursor
Anthropic API
Claude
Figma
"Prompt engineering"

We all know what it's like. People want to chat. People have questions. But sometimes people = many and Aidan = just one guy who occasionally needs to take a nice little snooze.

So, I built Aidan 2.0.

I designed it to look like iMessage because, well, familiarity is comforting. Plus it makes people feel like they're actually messaging with a real human (plot twist: they're not).

The technical bits: It's built in Django/Python with plain old HTML, CSS and JS, I used Cursor to help me code faster (AI helping me build AI me—very meta), crafted the front-end in Figma and spent approximately 347 hours fine-tuning the core prompt to make it sound just like me. Turns out teaching a bot to be dry without being mean is surprisingly difficult.

Project cards are formatted in a style similar to iMessage links

After v1, I discovered an unexpected feature: emptying my bank account. The API calls added up. So I added some preset questions and responses to keep costs down while maintaining the illusion of my digital presence.

The impact:

- I can now sleep through meetings (kidding... mostly)

- People get responses even when I'm busy making actual things

- I've learned more about how I communicate than any personality test could teach me

It turns out the best way to understand how you sound to others is to build a robot version of yourself and then realise it uses way too many em dashes—just like this.

The chatbot separates out it's messages into logical chunks and send multiple, so it's more like chatting to a real person